Description:The current invention is a NON-SMALL CELL LUNG CANCER TYPE AND STAGE DETECTION SYSTEM that automatically detects cancer type and stage.
BACKGROUND OF THE INVENTION
Although a significant amount of computer-assisted research has been done on Non-Small Cell Lung Cancer (NSCLC) diagnosis, there are still substantial challenges. Because NSCLC tumour exist in a variety of shapes and sizes, they're divided into solid, part-solid nodules, juxtavascular, juxtapleural, and pleural tail categories. Existing Computer-Aided Detection (CAD) systems, on the other hand, were limited to a specific nodule type, size, shape, and position. They were unable to deal with a variety of nodule shapes, sizes, locations, and heterogeneous intensity levels of low-dose CT images at the same time, resulting in erroneous positive and negative diagnoses. Many existing approaches are semi-automatic and do not perform in-depth classification of NSCLC types and stages according to the 8th edition TNM staging system, which physicians are now using in real time. Existing detection algorithms did not provide higher detection accuracy on a worldwide scale. Despite the extensive volume of research done in this field, no fully automated tool is currently available in hospitals/medical industry, which reflects the necessity for further research and development in the related technologies. Because of these factors, the development of a non-invasive "Fully Automated Non-Small Cell Lung Cancer Type and Stage Detection System" has been undertaken.
US20200370127A1- A US patent proposed methods and compositions for diagnosing or detecting a condition in a mammalian subject, such as lung disease, using a micro-RNA expression level or an expression level profile of multiple miRNA in the subject's peripheral blood mononuclear cells (PBMC), which is characteristic of COPD or NSCLC. Changes in expression of certain miRNA biomarkers from a reference sample or miRNA expression profile are linked to non-small cell lung cancer (NSCLC) and/or COPD, and can be used to distinguish between healthy people, those with COPD, and people with adenocarcinoma or squamous cell carcinoma.
CN106047998B- This application reveals the methods for detecting and using a specific lung cancer gene. The application's detection method includes cfDNA library construction and hybrid capture, upper machine sequencing of the library cfDNA using 12659 probes for the relevant 105 genes design of lung cancer, and data analysis of the sequencing result, resulting in comprehensive and accurate lung cancer gene information. The application's detection approach, which only requires a small sample of peripheral blood to complete, is really non-invasive. The application's detection method involves performing capture sequencing analysis for the lung cancer important gene in ctDNA, screening in the early stages of lung cancer, and providing information on medication associated gene mutations for patients with lung cancer. This provides an important reference frame for lung cancer early diagnosis, accurate medication guidance, pharmacodynamic assessment, and prognosis dynamic monitoring.
KR101075158B1- The current invention relates to a technique for detecting NSCLC using differentially expressed genes. In addition, the present invention introduces new human genes that have higher expression in non-small cell lung cancer tissue than in non-cancer tissue. The present invention additionally covers methods for finding chemicals that can be used to treat and prevent non-small cell lung cancer.

EP1737979B1-Methods for identifying non-small cell lung cancer (NSCLC) based on differentially expressed genes KIF11, GHSR1b, NTSR1, and FOXM1 are disclosed. Methods for discovering drugs for treating and preventing NSCLC based on the interaction of KOC1 and KIF11, or NMU and GHSR1b or NTSR1, are also described.
US9523130B2- In human solid malignancies, such as non-small cell lung carcinoma, novel gene deletions and translocations involving chromosome 2 have been reported, resulting in fusion proteins combining part of Anaplastic Lymphoma Kinase (ALK) kinase with part of a secondary protein (NSCLC). Echinoderm Microtubule-Associated Protein-Like 4 (EML-4) and TRK-Fusion Gene are two secondary proteins (TFG). It was confirmed that the EML4-ALK fusion protein, which preserves ALK tyrosine kinase activity, is responsible for the proliferation and survival of NSCLC with this mutation. Isolated polynucleotides and vectors encoding the disclosed mutant ALK kinase polypeptides, probes for detecting it, isolated mutant polypeptides, recombinant polypeptides, and reagents for detecting the fusion and shortened polypeptides are among the components of the invention. Ways for screening for chemicals that inhibit the proteins, as well as methods for stopping the course of a malignancy characterised by mutant polynucleotides or polypeptides, are now possible thanks to the discovery of this new fusion protein.
US7314721B2- Small cell lung cancer antigens and uses were proposed in a US patent. Autologous antibody screening of libraries of nucleic acids produced in small cell lung cancer cells using antisera from cancer patients discovered cancer related antigens. The invention pertains to nucleic acids and encoded polypeptides that are cancer-associated antigens expressed in small cell lung cancer patients. Isolated nucleic acid molecules, expression vectors containing those molecules, and host cells transfected with those molecules are all part of the innovation. Isolated proteins and peptides, antibodies to those proteins and peptides, and cytotoxic T cells that detect the proteins and peptides are also included in the invention. Additionally, functional fragments and variants of the foregoing are presented. Kits containing the aforementioned compounds are also available. The invention's molecules can be employed to diagnose, monitor, research, or treat illnesses characterised by the expression of one or more cancer-associated antigens.
CN104271159B The patent described a method for treating non-small cell lung cancer with TOR kinase inhibitors in a therapeutic alliance. The technique for treating or preventing advanced Non-small cell lung is described herein, which includes giving Erlotinib or a cytidine analogue of effective dose in combination with effective dose TOR kinase inhibitors to patients with advanced Non-small cell lung.

PRIOR ART SEARCH

KR101075158B1- method for diagnosing non-small cell lung cancers: 2005-07-18.
CN1854313B -Method for diagnosing non-small cell lung cancers: 2010-10-20.
EP1737979B1- Method for diagnosing non-small cell lung cancer: 2007-01-03.
US9523130B2- Methods of treating non-small cell lung carcinoma (NSCLC): 2016-11-17.
CN104271159B- The method that non-small cell lung cancer is treated using the therapeutic alliance of TOR kinase inhibitors: 2017-11-28.
US20200370127A1- Biomarkers in Peripheral Blood Mononuclear Cells for Diagnosing or Detecting Lung Cancers: 2020-04-30.
CN106047998B- A kind of detection method and application of lung cancer gene: 2016-10-26
JP6203209B2-Plasma microRNA for detection of early colorectal cancer: 2017-09-27.
US20200131586A1- Methods and compositions for diagnosing or detecting lung cancers: 2020-04-30.
US20200123613A1- Compositions and methods for diagnosing lung cancers using gene expression profiles: 2020-04-23.
US20210301350A1- Lung cancer determinations using mirna: 2021-09-30.
US20210147944A1- Methods for monitoring and treating prostate cancer: 2021-05-20.
US20210079479A1- Composition’s and methods for diagnosing lung cancers using gene expression profiles: 2021-03-18.
AU2011302344B2- HSP90 inhibitors for treating non-small cell lung cancers in wild-type EGFR and/or KRAS patients:
US7709202B2- Molecular characteristics of non-small cell lung cancer: 2010-05-04.
US7314721B2- Small cell lung cancer associated antigens and uses therefor: 2008-01-01.

REFERENCES:

1. Wong D.W-S. et al., "The EML4-ALK Fusion Gene is Involoved in Various Histologic Types of Lung Cancers from Nonsmokers with Wild-Type EGFR and KRAS", Cancer 115:1723-1733 (Apr. 2009).
2. Yamaguchi U. et al., "A Practical Approach to the Clinical Diagnosis of Ewing's Sarcoma/Primitive Neuroectodermal Tumour and Other Small Round Cell Tumours Sharing EWS Rearrangement Using New Fluorescence in situ Hybridisation Probes for EWSR1 on Formalin Fixed, Paraffin Wax Embedded Tissue", Journal of Clinical Pathology 58:1051-1056 (2005).
3. Takeuchi K. et al., "RET, ROS1 and ALK Fusions in Lung Cancer", Nature Medicine 18(3):378-381 (Feb. 12, 2012).
4. Soda M. et al., "Identification of the Transforming EML-4-ALK Fusion Gene in Non-small Lung Cancer", Nature 448:561-566 (2007).
5. Sasaki T. et al., "The Biology and Treatment of EML4-ALK Non-Small Cell Lung Cancer", European Journal of Cancer 46:1773-1780 (2010).
6. Oyama et al., Molecular Genetic Tumor Markers in Non-small Cell Lung Cancer:, Anticancer Research 25:1193-1196 (2005).
7. Ochs K. et al., "Case Report-Inflammatory Myofibroblastic Tumour of the Lung in a Five-Year-Old Girl", Interactive Cardiovascular and Thoracic Surgery 10:805-807 (2010).
8. Kwak E.L. et al., "Anaplastic Lymphoma Kinase Inhibition in Non-Small-Cell Lung Cancer", The New England Journal of Medicine 363(18):1693-1703 (Oct. 28, 2010).

OBJECTIVES OF THE INVENTION

• To design a Deep Learning based fully automated classification mechanism to facilitate precise and accurate detection of NSCLC tumours with their types and stages. The stage categorizer predicts the NSCLC’s stage by integrating the outputs from TNM state classifiers, as like as the latest eighth edition TNM staging system.
• To facilitate the medical practitioners in easier way to diagnosis NSCLC tumor by displaying the tumor region in RGB colored image rather than grayscale CT images.
• Provides a revolutionary fully automated smart prediction software tool that will save the doctor's initial disease screening time and boost the survival rate of patients.

SUMMARY OF THE INVENTION

The field of innovation is Computer-Aided Lung Cancer detection and categorization. Lung cancer has two main types. Non-Small Cell Lung Cancer (NSCLC) and Small Cell Lung Cancer (SCLC). We chose the NSCLC disease since, almost 87% of the total lung cancer is NSCLC, which affects both smokers and non-smokers. To discover malignant tumours, radiologists employ a variety of imaging techniques such as X-rays, magnetic resonance imaging (MRI), computed tomography (CT), and positron emission tomography (PET) scans. Despite being the best imaging tool in the medical profession, doctors find it difficult to interpret and diagnose NSCLC from CT scan data. Hence, it is essential to formulate a fully automated NSCLC diagnostic model to assist surgeons, in their precise diagnosis and for quick treatment planning.
Our innovative concept is to design a fully automatic categorization system using a Deep Learning classifier for NSCLC detection with its type and stage. A new post-processing scheme is introduced to enhance the nodule and critical features in the segmented image with RGB colors. Based on the GoogLeNet scheme, automatic classifiers (GoogLeNet-1, GoogLeNet-2, GoogLeNet-3, GoogLeNet-4, and GoogLeNet-5) are formed to achieve in-depth NSCLC classification. Initially, GoogLeNet-1 is produced to detect the NSCLC's presence. If the patient has cancer, then GoogLeNet-2 is facilitated to forecast the cell type for efficient treatment planning. To predict the T, N, and M states of images, the GoogLeNet-3, GoogLeNet-4, and GoogLeNet-5 classifiers are employed. The results of the TNM state classifier are gathered to discover stages based on the 8th edition of the TNM staging system. The proposed pre-processing, segmentation, and post-processing schemes considerably increase the accuracy of the proposed classification system. The proposed scheme will offer promising results as equivalent to experts’ decisions and present an efficient, fully automated prediction tool for experts to get well-timed and accurate diagnosis outcomes.

BRIEF DESCRIPTION OF THE INVENTION

Our invention, "A Fully Automated Non-Small Cell Lung Cancer Type and Stage Detection System," is based on a Deep Learning Classifier with color-coded CT scans. The proposed system's architecture is depicted in Figure 1. This system combines pre-processing and segmentation, post-processing, and classification using GoogLeNet to detect the existence, type, and stage prediction of NSCLC.
The presented pre-processing, segmentation, and post-processing method is intended to extract and assign RGB colour to the tumor region’s, lung wall’s, and mediastinum regions automatically. The Automated Multilevel Hybrid NSCLC Tumor Segmentation and Refinement Scheme was created to improve the accuracy of a fully automated classifier.

As a result, enhanced pre-processing processes are being proposed in order to correctly mine lung parenchyma structure, which is necessary for obtaining exact tumour segmentation and classification results. After mapping the grey intensity CT picture with RGB colours, either the B or G components are chosen and transformed back to grayscale, depending on which preserves more edge information. Then Contract Limited Adaptive Histogram Equalization (CLAHE), Wiener Filter (WF) and GrowCut (GC) algorithms are applied to isolate tumor region from lung wall an mediastinum region , and to eradicate frustrating background structures.
Followed by an Automatic Lung Parenchyma Mining and Border Restoration (ALPM&BR) scheme was applied, which included automatic seed point identification, region growing based mining, and new border concavity closing methods to attain impeccable mining that aids to isolate lungs by excluding the surrounding area. The nodules are mined using the Connected Component Analysis (CCA) and Threshold Based Mathematical Nodule (TBMN) processes, that efficiently eradicates irrelevant areas such as soft tissues, bone, vessels, fat, and so on.
The propounded scheme strives to achieve deeper classification like NSCLC type, TNM states, and predict stages using deep learning mechanism. The traditionally methods use either entire row CT image or segmented nodule patches to train the classification network. However, the NSCLC types [Adenocarcinoma, Squamous Cell Carcinoma and large cell carcinoma] categorization is based not only on its structural formation, but also it depends on the where the nodule’s located in the entire the lung region, such as near pleural region, mediastinum and/or isolated tumors. This information differs among all the patients. Instead of directly training the region with the network, here the classifier is designed to train with the images, which are subjected to pre-processing, segmentation, and post-processing techniques before the classifier is trained, to get precise results.
To emphasise the required image features and build a dataset suitable for type and TNM state categorization, the new post-processing scheme is required. The original image, lung parenchyma, and segmented nodule regions are the first three inputs obtained from the previous segmentation. The position of the tumour is crucial in determining the type, therefore the outside and inner lung walls, as well as tumour areas, are assigned various RGB colours in the original image and saved in the D1 database. Next, the NSCLC staging can be done from TNM state prediction process. Based on the Physician’s counsel, the ‘T’ state can be found from primary tumor size. The ‘T’ states are categorized into T1, T2, T3, and T4 based on the tumor sizes. There is not much variance in size amid neighboring ‘T’ states in grayscale, hence tumor region transformed into RGB green intensity and stored in the second database i.e. D2. Finally, the third database (D3) was created with the sole purpose of storing the mediastinum region taken from the original image. Because, the Nodule (N) state participation intensely depends on this area. The aforementioned newly designed post-processing scheme extensively support to attain accurate disease detection outcome as similar to the Physician’s detection process.
Next, the fully automated classification networks are created by Deep Learning technique, which is currently used in several image processing tasks efficiently with its self-characteristic learning proficiency. Here, there are five GoogLeNet classifiers are created named as GN1, GN2, GN3, GN4 and GN5 to perform preliminary classification (Non-cancer and cancer), NSCLC type detection, T state, N state and M state classification.
The following mechanisms are contributed in this classification work:
a) To accomplish a fully automatic NSCLC classifier, a new flow of pre-processing, automatic features segmentation, and post-processing is designed.
b) The experts perform preliminary investigation to analyze the non-cancer and cancer cases by observing the chief tumor size. According to the global cancer classification system, lung cancer is concluded if the chief tumor size is more than 0.5 cm. The database D2 which contains Pre-processed, Segmented, and Post-processed (PSP) images are taken to train and test the GoogLeNet-1 network with label 0 (non-Cancer) and 1 (cancer). The D2 comprise tumor images that have been transformed into RGB green intensity. This trained network now can capable to automatically predicts NSCLC presence if tumor size is greater than 0.5 cm
c) Deep learning enables accurate NSCLC types identification. The NSCLC has different types namely, Squamous cell carcinoma, large cell carcinoma, and adenocarcinoma. Using the DL scheme, an automated classifier (GoogLeNet-2) is produced and trained using the database D1 that contains PSP images in that, the nodule region, outer and inner lung wall are color mapped into green, red, and blue intensity.
d) Using the DL scheme, an automated classifier (GoogLeNet-3) is trained using the database D2, which comprise PSP images with tumor regions transformed into RGB green intensity. To forecast the 'T' state, the D2 dataset is divided into 'T' classes and trained using GoogLeNet-3. The major modification done in this work is the Pixel Classification Layer (PCL) employed, instead of the normal Classification Layer (CL), which efficiently identify the microscopic centimeter variance among the ‘T’ states. The models’ outputs belong to a non-binary class, thus the final output classification layer is updated with count 5 to predict the ‘T’ states that are identical to the expert's diagnosis. Currently, the experts fellows the 8th edition lung cancer classification system offered by the International Association for the Study of Lung Cancer (IASLC) to identify NSCLC cases. They done ‘T’ states classification based on the primary tumor size. Fundamentally, ‘T’ states include 5 classes namely, T0, T1, T2, T3, and T4, where ‘T0’ refers to the absence of primary tumor. The initial classifier (GoogLeNet-1) is used to filter out T0 cases. So that, here T0 in the ‘T’ state is not considered in the process of classification. State T1 is considered for tumors of size amid 0.5 and 3 cm, T2 is sub-divided into T2a with size band of 3 and 4 cm, and T2b with a size amid 4 and 5 cm, T3 with size from 5 to 7 cm, and T4 for size more than 7 cm. If a CT image includes several tumors, then the highest ‘T’ class is chosen.
e) The ‘N’ state shows whether and in what way the cells affect the lymph nodes in the neighboring area around the tumor. It has 4 classes like N0, N1, N2, and N3, where ‘N0’ represents that no regional lymph node is present, ‘N1’ denotes the ipsilateral pulmonary or hilar node metastasis, ‘N2’ indicates the subcarinal or ipsilateral mediastinal node metastasis and N3 signifies the contralateral mediastinal/hilar or supraclavicular node metastasis. It is tedious to model an automatic identification and classification system for ‘N’ states as the lymph node is positioned into the mediastinum area. The disparity and the structure of the abnormal lymph node are nearly similar to other organs. Hence, even the experts need heavy training and more knowledge to achieve effective classification. There are no semi-automatic or completely automatic 'N' state categorization systems in the literature because of the enormous difficulties involved in the process. With the trending DL's self-learning mechanism, the innovative proposed system easily achieves the fully automated 'N' state categorization system. In this case, GoogLeNet-4 is created and trained using the D3 database, which contains PSP images of the mediastinum alone. The DL model automatically learn the individual features of training images and map them to the predicted each class precisely.
f) f)‘M’ denotes the distant metastases showing that the cancer cells spread to another lung or other organs from the chief tumor. ‘M’ is classified into M0, M1a, M1b, and M1c, where ‘M0’ implies no remote metastases, ‘M1a’ denotes single or numerous tumors that spread to another lung, ‘M1b’ denotes the tumour that spread into an organ and ‘M1c’ denotes the tumor affected body parts or tumours spread into several organ. From the single axial view of the lung window image, M1b and M1c categories cannot be predicted. Hence, the GoogLeNet-5 classifier is trained with database D1 to forecast M0 and M1a classes.
g) The stage of cancer is forecast after getting the whole set of TNM states different from the prior research that simply find the stage only from the ‘T’ state result. The NSCLC stages include I (A, B), II(A, B), III(A, B, C), and IV(A, B). The lower stage (I) involves fewer transmitting risks and might be preserved entirely, while an advanced stage (IV) denotes the metastases. Once the results are gathered from GoogLeNet-3 (T state), GoogLeNet-4 (N state), and GoogLeNet-5 (M state) classifiers, the stages are predicted based on the current 8th edition TNM stage classification system.
, Claims:1. Fully automated NSCLC diagnosis system forecast aids to predict the patient’s survival rate.
2. According to claim 1, Offer an elite solution for automatic NSCLC Type detection which aids the Physician’s to plan an efficient treatment based on specific cell type.
3. According to claim 1, reduces patients' waiting time to receive a diagnosis report and affords enough time for treatment.
4. According to claim 1, Assist surgeons in their surgical planning by providing detailed information regarding: tumor structure formation, location, and area from 2D-CT scan image.
5. According to claim 1, Determining type and stage are vital as every type should be treated differently. To offer personalized treatment plans, patient upkeep, follow up and well-timed reclamation can be attained by the propounded fast and automatic forecast scheme.