Claims:Claims:
We Claim,

1. A Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet, wherein the process of hybrid model building is as follows:

Wherein the CT scan and X-ray images have been augmented by Image Data Generator Function of Keras;
Wherein further to make all input CT Scan or X-ray image sizes consistent, either padding process or standard loss less compression and feature extraction process is applied to get consistent image size of at least 256 x 256 pixel dimension;
Wherein hybrid model is based on convolutional neural network based architecture;
Wherein further the CNN feature extraction is carried out by multiple convo layers;
wherein at least 152 convo layers are used to build the Convolutional Neural Network of the CheXNet and CTxNet;
Wherein further the weights of the said convo layers are updated by binary cross entropy losses;
wherein in the present system, to do the learning more independently of each layer, the Batch normalisation is used to normalise the output of the previous layers;
Wherein the present system uses image-level prediction which uses small patches or segmented regions of an entire image for prediction of a category label (classification) or continuous value (regression) analysis;
Wherein the CT scan or X-ray Images were processed on a hybrid deep learning method based on at least one and more of the following parameters: Resize, Normalization, Rotation Range, Zoom Range, Cval, Shear Range, Horizontal Flip, and Vertical Flip;
Wherein the Input layer of image dimension is added on top of the CNN architecture, further Global Average Pooling Layer and one dropout layer are added, wherein the Output layer has seventeen neurons and sigmoid is further used for the activation of said seventeen output neurons;
wherein to analyse, detect and identify the Nodular Pattern the detection parameter is between 0.1 to 2.5;
Wherein the rib segmentation is used to analyse, detect and identify the cavity, bullas and bone pathologies for rib structure fracture analysis, which is processed by CLAHE (Contrast Limited Adaptive Histogram Equalization) Histogram Equalization for enhancing images;
Wherein for Pneumonia lobar and Segmental detection, the region-based segmentation is performed from left to right side and the zoom range is set between 0.5 to 5; and
Wherein for the filtration process, 256 to 500 filters for the RGB channel are used.
2. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, Wherein the present system detects either one and more of the following lung diseases using either X-ray image or CT scan images: Aortic enlargement, Atelectasis, Calcification, Cardiomegaly, ILD, Infiltration, Lung Opacity, Nodule/Mass, Other lesion, Pleural effusion, Pleural thickening, Pneumothorax, Pulmonary fibrosis, Covid, Edema, Pneumonia, Tuberculosis or normal condition by marking infected areas with heatmap.
3. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, wherein the present system identifies at least one and more of the lung infection, lung opacity, lung volume, bone density, or rib fractures on the chest CT scan or X-ray image.
4. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, Wherein the XChesNet model is used for X-ray image analysis and the CTxNet model is used for CT scan image analysis.
5. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, wherein the Global Average Pooling Layer with the dropout 0.16 (16%) value and dropout layer with a 0.3 (30%) value is added on top of the said CNN architecture.
6. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, wherein the Base Model Last Block Layer number is at least 126.
7. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, wherein the present disclosed novel deep learning system
achieves AUC as following, for identification of various diseases:
Calcification: 89% ± 0.2%, Cardiomegaly: 100% ± 0.2%, Interstitial lung disease (ILD): 71% ± 0.2%, Infiltration: 100% ± 0.2%, Lung Opacity: 88% ± 0.2%, Nodule/Mass: 92% ± 0.2%, Other lession: 82% ± 0.2%, Pleural effusion: 78% ± 0.2%, Pulmonary fibrosis: 96% ± 0.2%, Covid: 93% ± 0.2%, Edema: 92% ± 0.2%, Pneumonia: 93% ± 0.2%, Tuberculosis: 96% ± 0.2%, Aortic enlargement: 93% ± 0.2%, Atelectasis: 96% ± 0.2%, Cardiomegaly: 92% ± 0.2% and Normal: 100% ± 0.2%.
8. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, wherein for data preprocessing for the convo layers of the CT Scan and X Ray Images, the Resize of the said CT scan or X ray images is set to at least 256 x 256 x 3 pixel.
9. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, wherein for data preprocessing for the convo layers of the CT Scan and X Ray Images, the Normalization range of the said CT scan or X ray images is set to [ (0,255) -> (0,1) ].
10. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, wherein for data preprocessing for the convo layers of the CT Scan and X Ray Images, wherein the Rotation Range of the said CT scan or X ray images is set to (0, 0.05).
11. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, wherein for data preprocessing for the convo layers of the CT Scan and X Ray Images, wherein the Zoom Range of the said CT scan or X ray images is set to 0.05.
12. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, wherein for data preprocessing for the convo layers of the CT Scan and X Ray Images, wherein the Cval range of the said CT scan or X ray images is set to 0.05
13. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, wherein for data preprocessing for the convo layers of the CT Scan and X Ray Images, wherein the Shear Range of the said CT scan or X ray images is set to 0.05.
14. The Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet as claimed in claim 1, wherein for data preprocessing for the convo layers of the CT Scan and X Ray Images, wherein the Horizontal Flip and Vertical Flip of the said CT scan or X ray images is True.
, Description:Title of Invention
Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet

Field of Invention
The present novel invention relates to the technical field of lung disease identification by image processing, in particular to a method and a system for detecting R-CNN lung nodules by using CT scan images or X-ray images via improved three-dimensional image detecting system and method, which is based on Residual Networks (ResNet). In the present novel invention, the novel hybrid system is capable to process CT scan images or X-ray images. The present novel invention provides a classification and identification method for seventeen lung diseases, based on lung CT scan images or X-ray images via Residual Networks (ResNet) based novel deep learning method. In the present novel invention, a novel Hybrid ResNet based method is trained for identification of lung nodule, lung infection, lung opacity, lung volume, bone density and rib fractures. The present novel invention uses a hybrid deep learning approach to achieve promising accuracy above 90% on lung disease identification. Wherein the XChesNet is used for X-ray image analysis and the CTxNet model is used for CT scan image analysis.

Background of Invention
With the continuous deterioration of the environment caused by air pollution, the incidence and mortality of lung diseases are rising year by year. People are constantly exposed to particles, chemicals, and infectious organisms in ambient air, and the lung is the internal organ which is most vulnerable to infection and harm from the external environment. Respiratory disorders are a major cause of death and illness around the world. Five of these diseases are among the leading causes of death and severe illness in the world.
At present, the COVID-19 virus seriously threatens the health of human beings, the existing medical resources, diagnosis and treatment capacity are not enough to deal with the rapid rise of the diagnosis and treatment demand, and the personnel density in hospitals in a core epidemic area is increased, and the risk of cross infection is increased greatly.
At present, the method of CT scan and X-ray image scanning is based on the experience and judgement of the medical staff. Manual observation of the same is adopted for identifying lung disease in the prior art, the identification efficiency is low, and the method is directly related to the experience of the medical staff. At present since doctors in the hospitals judge lung nodule size based on experience by looking at visual inspection of the CT scan and X-ray images, a long time is needed for some interns to accumulate such experience. Also, the experience and technical level of each doctor is different, so there's a possibility of obtaining different results or outcomes. Also the visual method of inspection has certain subjectivity and is difficult to avoid error.
When using a CT scan apparatus to detect lung nodules of a patient, a radiologist needs to observe hundreds or even thousands of CT scan images one by one. This manual detection is not only time consuming but also brings a huge workload to the radiologist. Because doctors are easily fatigued in the process of reading the film, the working efficiency is affected, and the phenomena of missed examination and misdiagnosis occur.
Therefore, it is urgently needed to provide a method for identifying lung disease based on multi-task learning, which has simple logic and reliable identification.
Artificial intelligence deep learning techniques are increasingly applied in various fields due to rapid development in recent years, wherein a Convolutional Neural Network (CNN) model, which is one of the most important methods in deep learning techniques, has significant achievement in the aspects of classification, detection, segmentation and the same is true in the field of medical images. The convolutional neural network model is usually composed of a plurality of layers of neurons, so that the convolutional neural network model has strong characteristic learning capability, the learned network model has good representation capability on original data, internal rich characteristics of the data can be extracted through large-scale training data, and tasks such as data mining, detection and segmentation are favourably completed. Therefore, based on the related technologies such as the convolutional neural network, a large amount of data is used for training and learning image characteristics and carrying out models such as detection and segmentation.
At present, some existing methods for realising the volume of the lung nodule mainly judge the regional position of the lung nodule by calculating a first ROI (region of Interest) of each two-dimensional CT scan image, so as to calculate the size of the lung nodule. However, doctors are better at judging the malignancy of lung nodules through a two-dimensional model, and the sizes of the lung nodules obtained by the existing calculation method cannot bring intuitional feelings to the doctors.
Image segmentation, which is a fundamental problem in the field of image processing and computer vision, and in particular medical image segmentation, has recently become one of the most popular research directions in the field of medical images, and is of great significance for clinical diagnosis and treatment. In the aspect of target segmentation application of medical images, the conventional convolutional neural network structures mainly comprise U-Net, V-Net and the like.
Traditional pulmonary nodule detection is mainly divided into these steps: The first step is CT scan image pre-processing, then lung parenchyma extraction is carried out on the image with no noise, followed by extracting the suspected nodule region in the lung parenchyma, and finally, classifying the suspected nodule area, removing the pulmonary nodules in the suspected nodules, and reserving true positive nodules.
Problem and challenges of available technology and prior art:
Although traditional image processing based tests have achieved promising results, they still suffer from two significant drawbacks. The first drawback is that the conventional detection method is based on some simple assumptions and uses some low-level descriptive features. However, the shape, size, and texture of the true nodules in the lung have high variability, and low-level descriptive features cannot represent these true nodules, resulting in degradation of the overall detection result. The second disadvantage is that the detection method used by the traditional image processing detection method is generally divided into three sub-steps of lung segmentation, nodule candidate extraction, and false positive reduction. The whole detection process is long in time consumption, only suitable for small samples, incapable of being end-to-end, low in automation degree, and low in detection efficiency.
The existing target detection method for the novel coronavirus and other lung diseases, is generally performed based on a two-dimensional image, and therefore, the input for target detection performed in the field of medical images is generally a two-dimensional slice of a three-dimensional CT scan image, and a target (lesion region) is detected on the two-dimensional slice. This results in only the two-dimensional characteristics of the focus area being utilized in the detection process, and the three-dimensional structure information is not fully utilized so that the conditions of missed detection and false detection occur occasionally, and so the detection rate, the true positive rate and the accuracy rate of the detection cannot be ensured.
An invention disclosed in patent application number CN111899212A discloses a pulmonary nodule CT scan image detection method. The method comprises the steps of data preparation; generating a candidate nodule detection network; performing parameter adjustment processing on the generated candidate nodule detection network; training the candidate nodule detection network subjected to parameter adjustment processing by adopting a data set; and selecting optimal dropout and batch size values, and introducing a negative sample mining technology to optimise the data set. According to the method, pulmonary nodule parenchyma segmentation can be omitted, the Faster R-CNN basic network is changed, the network feature extraction capability is greatly improved, the improved multi-scale feature detection method also greatly improves the pulmonary nodule detection precision, and the method is of great significance to computer-aided diagnosis of pulmonary nodules.
An invention disclosed in patent application number CN112890768A discloses a lung distension disease identification method based on multi-task learning, and the method comprises the following steps: acquiring a plurality of lung images of patients with lung distension, patients with lung distension cured and healthy controls, carrying out a Gaussian filtering treatment on the acquired lung images, and marking the lung images of the patients with lung distension and the patients with lung distension cured; randomly dividing the lung images into a training set and a verification set; constructing a residual neural network (ResNet) of joint attributes; inputting the marked lung images in the training set and the lung images of the healthy controls into the ResNet, and obtaining the trained ResNet in combination with an attribute loss function; inputting the lung images of the verification set into the trained ResNet, and performing parameter optimization and adjustment to obtain an optimal ResNet; and acquiring lung images of patients, and inputting the acquired lung images to the optimal ResNet to identify whether the patients have lung distension disease or not.
An invention disclosed in patent application number CN112862824A discloses a novel coronavirus pneumonia focus detection method, system and device and a storage medium. The method comprises the following steps: acquiring a lung three-dimensional CT scan image of a patient; performing pre-processing operation on the CT scan image, wherein the pre-processing operation comprises lung region extraction, pixel normalisation and data enhancement; establishing a target detection network based on a three-dimensional image, wherein the detection network comprises a feature extraction network, a feature fusion network and a focus prediction network, the feature extraction network adopts 3D-Resnet as a basic framework, and the feature fusion network adopts the idea of 3D-FPN to perform feature fusion on feature layers of different scales, the focus prediction network adopts a Faster-RCNN target detection idea to carry out focus prediction on the fused feature layer; and during training or testing, inputting the processed image into the network to carry out novel coronavirus focus detection. The novel coronal pneumonia focus detection method can remarkably improve the detection rate, the true yang rate and the accuracy rate of a focus area.
The present novel invention Deep Learning Method to Diagnose Chest X-Ray or CT Scan Images based on Hybrid ResNet, utilises SE (Squeeze-and-excitation blocks) blocks to improve the representational capacity of a network by enabling it to perform dynamic channel-wise feature recalibration in order to facilitate doctors to diagnose more patients and start their treatment as soon as possible, as the present novel invention is also capable to diagnose early-stage infection, which is very difficult in the current scenario. Hospitals, small clinics, diagnosis centers, healthcare professionals can use the present novel system which can also be helpful in rural areas and villages. The present novel invention uses a hybrid deep learning approach to achieve promising accuracy above 90% on lung disease identification. The retrospective method has been used to train the present model. The standard model like the ResNet network that is publicly available has been modified significantly in the present novel invention. The present novel invention has at least a 121-layer convolutional neural network named XChesNet and CTxNet, where about 65,000 frontal view chest X-rays of seventeen different lung diseases were used to train in which XChesNet achieved F1 score of 0.435 where the score of radiologist was 0.387. The present novel system is designed as a refined framework to detect lung diseases using either X-ray image or CT scan image. The present novel invention does not pre-process the input image (either X-ray image or CT scan image) which is a normal step in the present conventional image processing methods, however in the present novel invention, If a CT scan or X-ray image will be of smaller size, then the padding process via the addition of zeros to the boundary of said images is performed. The present novel system also adds padding for the missing values so that the image can fit into the present prediction model. And if the said CT scan or X-ray image will be larger in size than the said dimensions, a standard lossless compression process will take place, to get the desirable optimized results.

Objectives of the Invention
? The principal objective of the present invention is to provide early detection of seventeen lung diseases using CT scans or X-ray images, that can effectively improve the survival quality and survival rate of the patients.
? Another objective of the present invention is to provide hybrid deep learning technology in the field of automatic analysis of medical images by virtue of high detection precision, high speed, no need of manually designing features, and the like.
? Another objective of the present invention is to use a hybrid deep learning approach to achieve promising accuracy above 90% on lung disease identification.
? Another objective of the present invention is to utilize SE (Squeeze-and-excitation) blocks to improve the representational capacity.
? The further objective of the present invention is to quickly and accurately detect the lung nodule on the chest CT scan or X-ray images.
? Another objective of the present invention is to provide computer-aided detection systems (CAD) to assist radiologists in detecting lung nodules.
? The further objective of the present invention is to bifurcate seventeen types of lung diseases and convert CT scans and X-ray images into 3D imaging for higher resolution for better identification and detection.
? Another objective of the present invention is to reduce human interventions, as imaging physicians repeatedly browse three-dimensional CT scan images layer by layer to find lung nodule areas and analyze the malignancy degree of lung nodules.
? Another objective of the present invention is that it does not pre-process the input image for its cropping or color-changing (either X-ray image or CT scan image), which are normal steps in the present conventional image processing methods.
? Another objective of the present invention is that, If a CT scan or X-ray image will be of smaller size, then the padding process via the addition of zeros to the boundary of said images is performed. The present novel system also adds padding for the missing values so that the image can fit into the present prediction model. And if the said CT scan or X-ray image will be larger in size than the said dimensions, a standard lossless compression process will take place, to get the desirable optimized results.
? Another objective of the present invention is to be used for the detection and notification of multiple diseases by analyzing chest X-ray or CT scan images, including tuberculosis, fracture, pleural effusion, and pneumothorax.
? The further objective of the present invention is to improve efficiency and reduce the rate of misdiagnosis.
? Another objective of the present invention is, it is able to be used easily at Hospitals, small clinics, diagnosis centers, healthcare professionals situated in cities, rural areas, and villages too.

Summary
Compared to the presently available technology and reported prior art, the present invention provides early detection of seventeen lung diseases using CT scans or X-ray images, that can effectively improve the survival quality and survival rate of the patients. The present novel invention provides hybrid deep learning technology in the field of automatic analysis of medical images by virtue of high detection precision, high speed, no need of manually designing features, and the like, it provides computer-aided detection systems (CAD) to assist radiologists in detecting lung nodules. The present invention bifurcates seventeen types of lung diseases and converts the CT scan and X-ray images into 3D imaging for higher resolution for better identification and detection. The present invention utilizes SE (Squeeze-and-excitation) blocks to improve the representational capacity. The present invention uses a hybrid deep learning-based approach to achieve promising accuracy above 90% on seventeen lung disease identification. The present novel invention quickly and accurately detects the lung nodule, it is also trained for identification of lung infection, lung opacity, lung volume, bone density, and rib fractures on the chest CT scan or X-ray images.
The present invention does not pre-process the input image (either X-ray image or CT scan image) which is a normal step in the present conventional image processing methods. If a CT scan or X-ray image will be of a smaller size, then the padding process via the addition of zeros to the boundary of said images is performed. The present novel system also adds padding for the missing values so that the image can fit into the present prediction model. And if the said CT scan or X-ray image will be larger in size than the said dimensions, a standard lossless compression process will take place, to get the desirable optimized results.
For the development of a present novel hybrid deep learning integrated interface to inspect machine-based Chest CT scan or X-ray images to inspect, identify and diagnose seventeen types of pathological labels and clinical findings also known/referred to as diseases, the following three datasets: NIH Chest X-rays 112k Dataset, NIH Chest CT32K Dataset, and VINBIG Chest Xray 18k Dataset were used for the machine learning purpose.
For the development of the present novel hybrid model, 50% images with known labels were used for model training, 40% images with known labels for testing, and 10% images with known labels for the validation, from the following datasets: NIH Chest X-rays 112k Dataset, NIH Chest CT32K Dataset, and VINBIG Chest Xray 18k Dataset.
In the present novel invention, applicants have developed a hybrid version of ResNet based on the neural network parameters, multiple convo layers, VGG neural network, and SE mechanism, the novel model is named as XChesNet and CTxNet. Wherein the VGG neural network is used for the image transformation. The said XChesNet and CTxNet based novel hybrid model detects seventeen diseases or infections from chest X-rays or CT scan images that work much more accurately than a well-practicing experienced radiologist, which is based on a novel hybrid deep learning integrated interface to inspect machine-based Chest CT scan or X-ray images. The XChesNet is used for X-ray image analysis and the CTxNet model is used for CT scan image analysis.
The present novel invention has the advantage that the CT scan or X-ray image to be detected is an input into the pulmonary nodule detection system, the image information can be automatically acquired and identified, and the lung parenchyma segmentation and the pulmonary nodule detection is sequentially carried out on the CT scan or X-ray images.
The lung parenchyma segmentation module greatly improves the segmentation precision of the lung parenchyma by introducing a residual error network structure and a boundary penalty item. The lung nodule candidate region detection module improves the detection precision of lung nodules by introducing a fast RCNN algorithm and eliminates false positive nodules by matching with a false positive nodule elimination module so that the detection precision of the lung nodules is further improved. And finally, outputting the detection result of the lung nodule through a lung nodule result output module, wherein the detection result comprises the size and the subsection information of the lung nodule. The system can automatically detect whether the CT scan or X-ray image has the nodules and the distribution condition of the nodules, and can improve the detection precision.

List of Figures
Figure 1: (a) The Block Chart of SE-ResNet and (b) Squeeze-and-excitation mechanism
Figure 2: Flow chart for present novel hybrid deep learning model building
Figure 3: The Confusion matrix of CT Scan and X-ray images
Figure 4: Comparison of X-Ray Image Original Label by Radiologist and Output Label by a present novel system
Figure 5: Comparison of CT Scan Images Original Label by Radiologist and Output Label by a present novel system
Figure 6: Precision, Recall, Classification category-wise for CT Scan and X-Ray images
Figure 7: Flow chart for CT Scan and X-Ray image analysis using a present novel model

Detailed Description of Invention
To further clarify the objects, technical solutions, and advantages of the present application, the present invention will be further described with reference to the accompanying drawings and examples, and embodiments of the present invention include but are not limited to, the following examples. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
It should be understood that this invention is not limited to the particular methodology, protocols, systems, etc., described herein and as such may vary. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of the present invention, which is defined solely by the claims. As used in the specification and appended claims, unless specified to the contrary, the following terms have the meaning indicated below:
“Architecture” refers to a set of rules and methods that describe the functionality, organization, and implementation of computer systems.
"Convolutional Neural Network (CNN)” refers to a class of deep, feed-forward artificial neural networks, most commonly applied to analyzing visual imagery. CNNs use a variation of multilayer perceptrons designed to require minimal preprocessing. A CNN consists of an input and an output layer, as well as multiple hidden layers. The hidden layers of a CNN typically consist of convolutional layers (also known as convo layers), pooling layers, fully connected layers, and normalization layers. Convolutional layers apply a convolution operation to the input, passing the result to the next layer. Local or global pooling layers combine the outputs of neuron clusters at one layer into a single neuron in the next layer. Fully connected layers connect every neuron in one layer to every neuron in another layer. CNNs use relatively little pre-processing compared to other image classification algorithms. This means that the network learns the filters that in traditional algorithms were hand-engineered. This independence from prior knowledge and human effort in feature design is a major advantage.
For example, machine learning analysis may receive and process images from medical imaging procedure data, to identify trained structures, conditions, and conditions within images of a particular study. The machine learning analysis may result in the automated detection, indication, or confirmation of certain medical conditions within the images, such as the detection of urgent or life-critical medical conditions, clinically serious abnormalities, and other key findings. Based on the result of the machine learning analysis, the medical evaluation for the images and the associated imaging procedure may be prioritized, or otherwise changed or modified. Further, the detection of the medical conditions may be used to assist the assignment of the medical imaging data to particular evaluators, the evaluation process for the medical imaging data, or implement other actions prior to, or concurrent with, the medical imaging evaluation (or the generation of a data item such as a report from such medical imaging evaluation).
Deep learning technology has been used for analyzing medical images in various fields in recent years, it shows excellent performance in various applications such as segmentation. The classical method of image segmentation is based on edge detection filters and several mathematical algorithms. Using several techniques improves targeted segmentation performance. To improve segmentation performance associated with medical images, DNNs and CNNs have been gradually introduced, but it is challenging to obtain high-quality, balanced datasets with labels in medical imaging. Medical images are mostly imbalanced, and time-consuming to obtain their labels. To overcome these issues, transfer learning is used. In transfer learning, a model trained on a large dataset is re-used and the weights determined in this model are applied to solve a problem that involves a small dataset. By transfer learning, one can also fine-tune weights via multiple experiments in some layers of the pre-trained model for effective and improved outcomes. For CNN, the earlier layers are typically frozen, as the last ones are freed up for tuning. This allows to perform full training on the existing model and modify the parameters at the very last layers. Transfer learning is an optimization, a shortcut to saving time or getting better performance. while using transfer learning, and weights are initialized with the strategy of “Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification”. Here, Kaiming initialization works well for the Neural Network where the ReLU activation function is used.
As further discussed herein, the machine learning analysis may be provided on behalf of any number of machine learning algorithms and trained models, including but not limited to deep learning models (also known as deep machine learning, or hierarchical models) that have been trained to perform an image recognition task, particularly for certain types of medical conditions upon medical images of human anatomy and anatomical representations. As used herein, the term “machine learning” is used to refer to the various classes of artificial intelligence algorithms and also rithm-driven approaches that are capable of performing machine-driven (e.g., computer-aided) identification of trained structures, with the term “deep learning” referring to a multiple-level operation of such machine learning algorithms using multiple levels of representation and abstraction. However, it will be apparent that the role of machine learning algorithms that are applied, used, and configured in the presently described medical imaging evaluation may be supplemented or substituted by any number of other algorithm-based approaches, including variations of artificial neural networks, learning-capable algorithms, trainable object classifications, and other artificial intelligence processing techniques.
The present novel hybrid deep learning integrated “XChesNet and CTxNet” based model inspects the Chest X-Ray or CT Scan images, which provides a facility to analyze datasets containing chest X-ray images or CT scan images for classification and segmentation purposes. Below are the diseases or infections which can be identified using a present novel hybrid deep learning integrated interface to inspect machine-based Chest CT scan or X-ray images:
1. Aortic enlargement
2. Atelectasis
3. Calcification
4. Cardiomegaly
5. ILD
6. Infiltration
7. Lung Opacity
8. Nodule/Mass
9. Other lesion
10. Pleural effusion
11. Pleural thickening
12. Pneumothorax
13. Pulmonary fibrosis
14. Covid
15. Edema
16. Pneumonia
17. Tuberculosis

The present novel model is also capable of identifying that the Chest CT scan or X-ray images of the patient are in normal condition.
Further detailed description and working of the present novel invention for development of novel Hybrid Deep Learning Method for Detection of Critical Findings in Lung using Chest X-ray or CT scan Images based on ResNet, is given in the following flow along with the form of examples and detailed description:
1.1 Datasets
1.1.1 NIH Chest X-rays 112k Dataset
1.1.2 NIH Chest CT32K Dataset
1.1.3 VINBIG Chest Xray 18k Dataset
1.2 Reading the Scans
1.3 Developing the Hybrid Deep Learning Novel Model
1.3.1 Lung Segmentation (AOI)
1.3.2 Preprocessing
1.3.3 Analysing Pattern by CNN Feature Extractor
1.3.4 XChesNet and CTxNet Model Building
1.3.5 Model Building, Implementation and Training
1.3.6 Evaluating and Comparison of novel model to Radiologist and its Statistical Analysis
1.3.7 Comparison of Novel Model Outcome with Experienced Radiologist
1.4 Results and Conclusion
1.1 Datasets
For the development of a present novel hybrid deep learning integrated interface to inspect machine-based Chest CT scan or X-ray images to inspect, identify and diagnose seventeen types of pathological labels and clinical findings, the present novel system has used the following three datasets: NIH Chest X-rays 112k Dataset, NIH Chest CT32K Dataset, and VINBIG Chest Xray 18k Dataset.
1.1.1 NIH Chest X-rays 112k Dataset
National Institute of Health (NIH), Chest X-ray Dataset comprises 112,120 X-ray images with seventeen disease labels. In total, 108,948 frontal-view X-ray images are in the database, of which 24,636 images contain one or more pathologies. The remaining 84,312 images are normal cases. The main body of each chest X-ray report is generally structured as “Comparison”, “Indication”, “Findings”, and “Impression” sections. The said dataset is available on Kaggle platform at: https://www.kaggle.com/nih-chest-xrays/data, which the present novel system has used for the machine learning of the present novel invention. In the present invention, applicants have used this dataset for novel model building and for training purposes, also in the present novel invention development, applicants have used about 30k unique images from the said dataset for validation and testing of the novel developed algorithm. The present invention focuses on detecting disease concepts in the findings and impression sections.
1.1.2 NIH Chest CT32K Dataset
The National Institutes of Health’s Clinical Center has made a large-scale dataset of CT images publicly available to help the scientific community improve the detection accuracy of lesions. This dataset by NIH, named DeepLesion, has over 32,000 annotated lesions identified on CT images. The images, which have been thoroughly anonymized, represent 4,400 unique patients, who are partners in research at the NIH. The said dataset is available on Kaggle platform at: https://www.kaggle.com/kmader/nih-deeplesion-subset which the applicants have used for the machine learning and model building of the present novel invention. The dataset released is large enough to train a deep neural network – to develop a universal lesion detector that is helping radiologists to find all types of lesions. Based on this the present novel invention is developed to create a large-scale universal lesion detector with one unified framework. The present novel invention is capable of more accurately and automatically measuring sizes of all lesions a patient may have, enabling the whole body assessment in an easy, accurate, and efficient manner.
1.1.3 VINBIG Chest Xray 18k Dataset
The present invention uses preprocessed images from VinBigData. To the said dataset the original Dicom format is converted to a png image, preserving the resolution and aspect ratio. The Preprocess algorithm helps to augment the images, which are uploaded in the kernel processing, having the original size and lossless png. The VinDr-CXR dataset is built to provide a large dataset of chest X-ray (CXR) images with high-quality labels for the research community, this dataset has more than 100,000 raw images in DICOM format that were retrospectively collected from 108 Hospitals and the Hanoi Medical University Hospital, from Vietnam. The published dataset consists of 18,000 posteroanterior (PA) view CXR scans that come with both the localization of critical findings and the classification of common thoracic diseases. These images were annotated by a group of 17 radiologists with at least 8 years of experience for the presence of 22 critical findings (local labels) and 6 diagnoses (global labels); each finding is localized with a bounding box. The local and global labels correspond to the “Findings” and “Impressions” sections, respectively, of a standard radiology report. This dataset is divided into two parts: the training set of about 15,000 scans and the test set of about 3,000 scans. The said dataset is available on Kaggle platform at: https://www.kaggle.com/corochann/vinbigdata-chest-xray-original-png, which the present novel system has used for the machine learning and model building of the present novel invention.

1.2 Reading the Scans
For the development of the present novel hybrid model, about 50% of images with known labels from above-mentioned dataset of Chest CT scan images and X-ray images from various sources (NIH Chest X-rays 112k Dataset, NIH Chest CT32K Dataset, and VINBIG Chest Xray 18k Dataset) were imported to the dataset. The said dataset has measurement and markings for clinically meaningful findings by the experienced radiologists, by an electronic bookmark tool, and was used to make the system learn and for model training.
Random images and reports of the Chest CT scan and X-ray from the said dataset are further evaluated by a group of experienced radiologists to check the appropriateness of the said dataset.
For the compilation of master dataset preparation for the development of present novel hybrid deep learning integrated interface to inspect machine-based Chest CT scan or X-ray images to inspect, identify and diagnose seventeen types of pathological labels and clinical findings, the findings, and bookmarks mentioned in said clinical reports written by radiologists were considered as the gold standard.
The present novel hybrid deep learning integrated interface was tested using 40% images with known labels and validated by 10% images with known labels based on the reports from the subset of said datasets (NIH Chest X-rays 112k Dataset, NIH Chest CT32K Dataset, and VINBIG Chest Xray 18k Dataset) to ensure that the inferred information was accurate and could be used as the gold standard.

1.3 Developing the Hybrid Deep Learning Novel Model
Figure 2 represents the flow chart for the XChesNet and CTxNet based for present novel hybrid deep learning model building, the detailed description of the same is given in following paragraphs:
Deep learning is a form of machine learning where the model used is a neural network with a large number of (usually convolutional) layers. Training this model requires a large amount of data for which the truth is already known. Training is usually performed by an algorithm called backpropagation. In this algorithm, the model is iteratively modified to minimize the error between predictions of the model and the known ground truth for each data point.
In the present novel invention, applicants have developed a hybrid version of ResNet, which is based on multiple neural network parameters, multiple convo layers, VGG neural network. The novel model is named as XChesNet and CTxNet. Wherein the VGG neural network is used for the image transformation. The said XChesNet and CTxNet based model detects seventeen diseases or infections from chest X-rays or CT scan images that work much more accurately than a well-practicing radiologist, which is based on a novel hybrid deep learning integrated interface to inspect machine-based Chest CT scan or X-ray images. XChesNet is used for X-ray image analysis and the CTxNet model is used for CT scan image analysis.
XChesNet and CTxNet are deep Convolutional Neural Networks consisting of at least 152 layers. The novel hybrid deep learning-based model produces a heatmap that localizes the areas in which disease symptoms are highly indicative in the CT scan or X-ray images along with the prediction probability. The architecture of the novel XChesNet and CTxNet based model includes all convolutional layers and the fully connected layers.
For the XChesNet and CTxNet model building of the present novel invention Transfer Learning approach is used, wherein the base model is SE-ResNet. imageNet is used to initialize the weights of one and more of the above-mentioned convo layers. On the trail to this ResNet network, global pooling, with sigmoid function along with dropout layers are attached to predict seventeen diseases or infections.
In the present novel invention, Deep Learning Method to Diagnose Chest X-ray or CT scan Images based on Hybrid ResNet is also trained for identification of Lung infection, lung opacity, lung volume, bone density, and rib fractures.
1.3.1 Lung Segmentation (AOI)
Before training, CT scan and X-ray images have been augmented by Image Data Generator Function of Keras, where the augmentation is carried out based on one and more of the factors like as to: rotation range, shear range, zoom range, cval, horizontal flip, vertical flip, rescale, whitening, width shift range, height shift range, and channel shift range has been used to remove other regions like as to shoulder part.
1.3.2 Preprocessing
The classification or regression model is used for the prediction of missing values depending on the nature of the feature having missing value. For the present novel invention, CT scan and X-ray image are taken as input. Before passing the CT scan and X-ray images to CNN Layer, it is essential to have all image sizes consistent of at least 256 x 256 Dimension. If a CT scan or X-ray image will be of smaller size, then the padding process via the addition of zeros to the boundary of said images is performed. The present system also adds padding for the missing values so that the image can fit into the present prediction model. And if the said CT scan or X-ray image will be larger in size than the said dimensions, a standard loss-less compression process will take place. Likewise, a model based on similar fundamentals and processes with all image sizes with any other dimensions can be developed and on need, the input images can be processed via padding process or lossless compression process.
1.3.3 Analysing Pattern by CNN Feature Extractor
Once the CT scan and X-ray images augmentation and processing are carried out, further they are processed for analyzing patterns by CNN feature extractor. The CNN feature extraction is carried out with the help of multiple convo layers. At least 152 convo layers are used to build Neural Networks. CT scan and X-ray images are surpassed to these layers where feature extraction takes place with this convo layer. Further one and more of the MaxPooling, Global Average Pooling, Dropout layers, batch normalisation, ReLu correction layer are added to tune training accuracy with one and more of the ReLU, sigmoid and softmax activations.
1.3.4 XChesNet and CTxNet Model Building:
XChesNet is used for X-ray image analysis and CTxNet model is used for CT scan image analysis. The said XChesNet and CTxNet based model detects seventeen diseases or infections.
For the machine learning of the present novel invention, chest CT scan and chest X-ray images are incorporated for the classification of the dataset, that consists of data with two classes–normal (healthy person) and those containing marks of either one and more of seventeen listed diseases dataset as obtained from the above mentioned listed datasets.
In the present novel invention, the Batch normalization allows each layer to do the learning more independently and is used to normalize the output of the previous layers. Activations scale the input layer in normalization. After using batch normalization learning became more efficient, which is also used as a regularisation to avoid overfitting of the model. It is added to the sequential model to standardize the input or the outputs.
Also, Random lateral inversion was applied with at least 50% probability before being fed into the convo layers. Apart from this, weights of the convo layer are updated by binary cross-entropy losses.
In this work, the present novel system uses the term ‘image-level prediction’ to refer to tasks where the prediction of a category label (classification) or continuous value (regression) is implemented by analysis of an entire image. These methods are distinct from those which make predictions regarding small patches or segmented regions of an image. Classification and regression tasks are grouped together in the present novel work, since they typically use the same types of architecture, differing only in the final output layer.
Improvement of model-based predictions occurs with respect to the increasing number of training datasets. To synthetically augment the training dataset, some transformations have to be applied to the existing prior art dataset in which, pre-processing is easy to use in these scenarios but it affects the original image as well as prediction, therefore the present novel invention does not pre-process the input image (either X-ray image or CT scan image), however cropping and lossless compression of an image is done to meet the requirements of the present system to get the desirable results.
In the present novel invention, the Deep Neural Layers contain at least 152 convo layers, out of which the first convo layer is also known as an input layer, wherein the RGB image is not directly inserted into the convo layer, it is first resized into pixel dimension of at least 256 X 256.
Figure 1 (a) represents the Block Chart of SE-ResNet. In the SE-ResNet Model, a squeeze and excitation (SE) mechanism (Figure 1 (b)) is attached with ResNet which forms SE-ResNet. Herein the output of the ResNet Neural Network is supplied to the SE mechanism. In all, the CT scan and X-ray images are first converted to tensor and passed to the next convo block by transforming it with the help of filters supplied on layers. Transformations like squeezing, excitation, and scaling are performed.
Squeezing is global average pooling, Excitation is a layer, where tensor is transformed to matrix linearly, then non-linearity is performed and then output is applied to sigmoid, to give a number either 0 or 1.
SE (Squeeze-and-excitation) mechanism improves the representational capacity of a convolutional neural network by enabling it to perform dynamic channel-wise feature recalibration.
After SE, the "OR" operation will be performed with the initial tensor and SE produced output, which is fed as backpropagation. Like this whole layer is designed and trained.
Also, Random lateral inversion is applied with 50% probability before being fed into the ResNet. Apart from this, weights of the neurons of the convo layers are updated by binary cross-entropy losses.
wherein the SE (Squeeze-and-excitation) Block is calculated via:

Equation 1 (SE Block Formula)

Wherein the Ftr (residual block) is the convolutional operator for the transformation of X to U, and is calculated via:

Equation 2 (2D spatial kernel Formula)
Where,

Ftr = Transformation
X = Input
U = Set of Feature Maps (Output)
V = Set of filters kernels
R = Relation
H x W = Spatial Dimensions
C = Convo Layer
H’ x W’ = Input (previous) Spatial Dimensions
C’ = Previous Convo Layer
* = Convolution
uc = Output to c-th filter
vc = Parameter refer to c-th filter

1.3.5 Model Building, Implementation, and Training
For training the model, the present novel system has used 50% images with known labels from the following datasets: NIH Chest X-rays 112k Dataset, NIH Chest CT32K Dataset and VINBIG Chest Xray 18k Dataset.
For training the model, png and jpeg images of a unified size of at least 256 X 265 pixels were used. The transfer learning approach is used where our base model is Se-ResNet which is based on XChesNet and CTxNet with at least 152 convo layers.
CT scan and X-ray Images were processed on a novel hybrid deep learning method based on the following parameters:
Methods Setting
Resize 256 x 256 x 3 (min pixel size)
Normalization [ (0,255) -> (0,1) ]
Rotation Range (0, 0.05)
Zoom Range 0.05 range
Cval 0.05 range
Shear Range 0.05 range
Horizontal Flip True
Vertical Flip True
Table 1: Data Pre-processing Parameters For Various Convo Layers
The present novel XChesNet and CTxNet is built on top of Se-Resnet, which has at least 152 convo layers. Further tuning the hyperparameter for one or more of the convo layers on Keras and TensorFlow framework is performed. As the Se-ResNet model has been taken as transfer learning and it has been tuned and added layers according to need in which ‘imagenet’ weights have been used to initialize the weights to neurons of the convo layer. The Input layer of image dimension is added on top of CNN architecture. Further Global Average Pooling Layer is added with the dropout 0.16 (16%) value and one dropout layer is added with a 0.3 (30%) value. The Output layer has seventeen neurons which are used to identify above mentioned seventeen diseases or ailments, sigmoid is used for the activation of said seventeen output neurons. Base Model Last Block Layer number refers to the number of layers in the base model used to train deep learning. In the present system, the Base Model Last Block Layer number is at least 126.
During training, ReduceLROnPlateau, ModelCheckpoint, EarlyStopping as callbacks have been used which monitor training accuracy.
For the training purpose, the present novel system has an annotation dataset that consists of the above-mentioned seventeen lung diseases. The parameter of Nodular Pattern detection is between 0.1 to 2.5 through this the present novel system analyzed, detected, and identified the nodular pattern. After that the novel system analyses for identification of lung infection, lung opacity, lung volume, bone density and rib fractures. In the rib structure fracture analysis, the novel system analyses to detect and identify the bone density, for this the system uses the rib segmentation which is processed by CLAHE (Contrast Limited Adaptive Histogram Equalization) Histogram Equalization for enhancing images and through the above-mentioned Lung Segmentation method. For Pneumonia lobar and Segmental detection, the present novel system used the region-based segmentation for left to right side and the zoom range is between 0.5 to 5. For the filtration process, the present novel system used 256-500 filters for the RGB channel (Red, Green, Blue).
The present novel invention provides the RGB images for the black and white input images of CT scan and X-ray images. Said RGB images represent the heat map to show and represent the infected areas of the lung towards various disease conditions as shown Figure 4 and 5.
The said Figure 4 and 5, shows indicative images for RGB images for the black and white input images of CT scan and X-ray images for various disease conditions, similarly for all disease conditions and all input chest X-ray or CT scan images, the present novel system gives an output of RGB images.
1.3.6 Evaluating and Comparison of Algorithms to Radiologist and its Statistical Analysis

For evaluating the present novel system for identification of seventeen lung diseases from the CT scan or X-ray images, based on the novel XChesNet and CTxNet Model built on above-mentioned data pre-processing parameters on the said images, the novel system has used 40% images for testing and 10% images for the validation, with known labels from the following datasets: NIH Chest X-rays 112k Dataset, NIH Chest CT32K Dataset, and VINBIG Chest Xray 18k Dataset.
The results of the said novel system for prediction of seventeen lung diseases identification from the said know labeled CT scan and X-ray images were further analyzed for precision, sensitivity or recall, F1 score, and accuracy using the following method and formulas:
Precision Calculation:

Equation 3: Precision Calculation

Sensitivity or Recall Calculation:

Equation 4: Recall calculation

Accuracy Calculation:

Equation 5: Accuracy Calculation

F1 Score Calculation:

Equation 6: F1 Score Calculation

Where,
TP = True Positive
FP = False Positive
FN = False Negative
TN = True Negative
FN = False Negative
The results of the present novel system for precision, sensitivity or recall, F1 score and accuracy analysis for the chest X-ray or CT scan images processed through present novel XChesNet and CTxNet based deep learning for the prediction of seventeen lung diseases identification is as follows:
Accuracy Precision Sensitivity or Recall F1 Score
X-ray 91% 98.2% 94.6% 96.4%
CT scan 98% 97.5% 98% 97.8%

Table 2: Accuracy, Precision, Recall, and F1 score analysis
The overall accuracy of the present novel invention to diagnose chest X-ray images for seventeen lung diseases is 91%, 98.2% precision and 94.6% sensitivity; and the overall accuracy of the present novel invention to diagnose chest CT scan images for seventeen lung diseases is 98%, 97.5% precision and 98% sensitivity, which will gradually increase by increasing the dataset.
A confusion metric is a performance measurement technique for machine learning classification problems. By calculating a confusion matrix, it gives details related to Machine Learning Model such as models’ accuracy, and also provides information of false prediction by model.
Figure 3 represents the Confusion matrix of CT scan and X- ray images for identification of seventeen lung diseases, based on the 10% images used for validation of the present novel system, with known labels from the following datasets: NIH Chest X-rays 112k Dataset, NIH Chest CT32K Dataset and VINBIG Chest Xray 18k Dataset.

1.3.7 Comparison of Novel Model Outcome with Experienced Radiologist
To compare and validate the performance of the present novel trained system, for deep learning-based method to diagnose chest X-ray or CT scan Images based on hybrid ResNet model building, one thousand images each of X-ray scans or CT scans were processed through the present novel system and experienced radiologists, and the results were shown in Figure 4 and Figure 5.
Figure 4 represents the comparison of X-ray images original label by radiologist and output label by a present novel system
Figure 5 represents the comparison of CT scan images original label by radiologist and output label by a present novel system
The said Figure 4 and 5, shows indicative images for RGB images for the black and white input images of chest CT scan or X-ray images for various disease conditions, similarly for all disease conditions and all input chest X-ray or CT scan images, the present novel system will give an output of RGB images.

1.4 Result and Conclusion
The present disclosed novel deep learning system demonstrated an AUC of the following range for the various lung diseases identification from the input CT scan and X-ray images:
Accuracy of Calcification: 89% ± 0.2%, Cardiomegaly: 100% ± 0.2%, Interstitial lung disease (ILD): 71% ± 0.2%, Infiltration: 100% ± 0.2%, Lung Opacity: 88% ± 0.2%, Nodule/Mass: 92% ± 0.2%, Other lession: 82% ± 0.2%, Pleural effusion: 78% ± 0.2%, Pulmonary fibrosis: 96% ± 0.2%, Covid: 93% ± 0.2%, Edema: 92% ± 0.2%, Pneumonia: 93% ± 0.2%, Tuberculosis: 96% ± 0.2%, Aortic enlargement: 93% ± 0.2%, Atelectasis: 96% ± 0.2%, Cardiomegaly: 92% ± 0.2% and Normal: 100% ± 0.2%
After checking the performance comparison between the present novel trained system and the experienced radiologists for CT scan images and X-ray images, it is further analyzed from the validation images dataset (10% images with known labels from the following datasets: NIH Chest X-rays 112k Dataset, NIH Chest CT32K Dataset and VINBIG Chest Xray 18k Dataset). For the cross-validation analysis of the present novel invention, from the said dataset the unlabelled 2550 X-ray images and 2000 CT scan images were used for further analysis out of 10% of total validation images.
The following analysis report has been generated, which shows precision and sensitivity analysis per category for the said image dataset:
2550 X-ray images 2000 CT scan images
Category Precision Sensitivity Precision Sensitivity
Covid 0.99 0.93 0.95 0.98
Edema 0.96 1.00 0.68 0.97
Pneumonia 0.99 0.99 0.98 0.99
Tuberculosis 1.00 0.99 1.00 1.00
Aortic enlargement 0.94 0.93 0.85 0.91
Atelectasis 0.92 0.90 0.89 0.90
Cardiomegaly 0.90 0.91 0.90 0.91
Lung Opacity 0.92 0.90 0.94 0.95
Calcification 0.89 0.95 0.95 0.92
Cardiomegaly 0.92 0.96 0.91 0.94
Interstitial lung disease (ILD) 0.95 0.94 0.94 0.94
Infiltration 0.95 0.98 0.96 0.90
Nodule/Mass 0.97 0.95 0.92 0.98
Pleural effusion 0.95 0.94 0.90 0.98
Pulmonary fibrosis 0.94 0.98 0.91 0.94
Other lesson 0.85 0.90 0.80 0.85
Normal 0.97 1.00 1.00 1.00

Table 3: Disease wise precision and sensitivity of overall system

To determine whether XChesNet and CTxNet performance are statistically significantly higher than radiologist performance, the present novel system also calculates the difference between the average F1 score of XChesNet and CTxNet and the average F1 score of the radiologists on the same bootstrap samples. If the 95% clearance on the difference does not include zero, the present novel system concludes there was a significant difference between the F1 score of XChesNet and CTxNet and the F1 score of the radiologists. The present novel system finds that the difference in F1 scores — 0.051 (95% Clearance 0.005, 0.084) — does not contain 0, and therefore concludes that the performance of present XChesNet and CTxNet based novel model is statistically significantly equivalent to that of the experienced radiologist performance.
The accuracy of the present novel invention for chest CT scan and X-ray image scanning is high and the rate of False Negative prediction is very low, which indicates that the present novel model performance is good. The overall accuracy of the present system is 91% which will gradually increase as the dataset increases.
When X-ray or CT-Scan image is fed into XChesNet model, it undergoes from many convo layers where neurons of each layer get activated and prediction is carried out. As proposed, CNN will diagnose 17 diseases and whichever get higher prediction after passing through sigmoid activation will be listed in final detection list.

Example
A set of multiple chest X-ray or CT scan images were processed through herewith disclosed novel Deep Learning Method for identification of various lung disease conditions, and the results obtained are disclosed in the following table:
Category AUC of X-ray and
CT scan High sensitivity
operating point High specificity
operating point
Covid 93% ± 0.2%,
0.93 0.98
Edema 92% ± 0.2%,
1.00 0.97
Pneumonia 93% ± 0.2%
0.99 0.99
Tuberculosis 96% ± 0.2%
0.99 1.00
Aortic enlargement 93% ± 0.2%
0.93 0.91
Atelectasis 96% ± 0.2%
0.90 0.90
Cardiomegaly 92% ± 0.2%
0.91 0.91
Lung Opacity 88% ± 0.2%
0.90 0.95
Calcification 89% ± 0.2%
0.95 0.92
Cardiomegaly 100% ± 0.2%
0.96 0.94
Interstitial lung disease (ILD) 71% ± 0.2%
0.94 0.94
Infiltration 100% ± 0.2%
0.98 0.90
Nodule/Mass 92% ± 0.2%
0.95 0.98
Pleural effusion 78% ± 0.2%
0.94 0.98
Pulmonary fibrosis 96% ± 0.2%
0.98 0.94
Other lesson 82% ± 0.2%
0.90 0.85
Normal 100% ± 0.2%
1.00 1.00

Detailed process for CT Scan and X-ray image analysis using the present novel method:

Figure 7 represents the Flow chart for CT Scan and X Ray image analysis using present novel model, it represents the following step based process:

Step 1: Upload the machine-generated chest X-ray or CT scan image in a given format either from .jpg , .jpeg, .DICOM or .png file in the present novel system.
Step 2: Further the present novel system first checks that image either belongs to X-ray or CT scan based on the pixel size of the image. Preferably it should be at least 256 x 256 x 3 pixels and above, and if required the same is achieved via padding process or standard lossless compression process to have all images of consistent size.
Step 3: The resizing of the said input X-ray or CT scan image is performed by VGG Neural Network based on one and more of the following parameters: Resize, Normalization, Rotation Range, Zoom Range, Cval, Shear Range, rescale, whitening, Horizontal Flip and Vertical Flip
Step 4: Further the present novel system checks, If the image does not belong to X-ray or CT scan image then it will reject the image and notify the user to upload X-ray or CT scan images, else it will continue to feed the image in present novel XChesNet and CTxNet based deep CNN model.
Step 5: Further the said chest X-ray or CT scan image is processed for Batch normalization by present novel XChesNet and CTxNet based deep CNN model.
Step 6: Further present novel XChesNet and CTxNet deep CNN model applies the hyperparameter tuning for the analysis and bifurcation of the images for identification of seventeen lung diseases. One and more of the following hyperparameters were applied: Optimizer, Learning Rate, Learning Rate Decay Per Epoch, Batch Size, Hidden Layer Activation Function, Classification Activation Function, and Loss.
Step 7: Based on the said one and more hyperparameters, the score value for SE (Squeeze-and-excitation) Block for each of the input chest X-ray or CT scan images is calculated based on the following equation:
SE (Squeeze-and-excitation) Block is calculated via:

Equation 1: SE Block Formula
Step 8: Further for each of the said input chest X-ray or CT scan images, based on the above-derived score value via SE Block Formula, class is further defined in the range of 0 to 16 via the 2D spatial kernel Formula, to identify one and more of the listed seventeen lung diseases.

Equation 2: 2D spatial kernel Formula
Step 9: Further it will under-go for a heat map generation with respect to the diagnosed disease, and the uploaded image will be plotted with the color of the RGB channel which shows the high and low infection areas.
Step 10: After analyzing and predicting the chest X-ray or CT scan images, the present novel system will return results to the user by marking infected areas with heatmap along with a proper diagnostic result message, where the user can further download the said report.