Description:Form 2

THE PATENT ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003
COMPLETE SPECIFICATION
(See section 10 and rule 13)

“A SYSTEM FOR LUNG CANCER SEGMENTATION USING DEEP LEARNING TECHNIQUES”

in the name of UNIVERSITY OF MADRAS an Indian National having address at UNIVERSITY OF MADRAS, CHEPAUK, CHENNAI, CHENNAI – 600005, TAMIL NADU, INDIA.

The following specification particularly describes the invention and the manner in which it is to be performed.

FIELD OF THE INVENTION:

The present invention generally relates to the field of biomedical engineering and medical imaging. More particularly, the present invention relates a system for lung cancer segmentation using deep learning techniques.

BACKGROUND OF THE INVENTION:

Lung cancer is one of the most mortality cancers at present, and the definitive diagnosis of cancer requires localization of the cancerous region first and then type discrimination. Among them, histopathological image analysis can be used as a gold standard for lung cancer diagnosis.

Classification and assessment of the extent of cancer types is crucial for targeted therapy. In clinical practice, experienced pathologists identify cancer by scanning H & E stained tissue slides into full-field digital sections and observing a definitive diagnosis, which is a time consuming and laborious task due to the extremely large size of the Image data, normal areas are relatively similar to cancerous areas, e.g. requiring an experienced histopathologist examination for about 15 minutes to half an hour to examine a complete section. Therefore, Computer Aided Diagnosis (CAD) systems place high demands on automated analysis techniques in the field of pathology, which can greatly reduce the workload and speed up the Diagnosis to help timely treatment.
There are reports available in the state of art revealing the existence of systems for lung cancer segmentation.

CN111709929A discloses a lung canceration region segmentation and classification detection system, which comprises the steps of firstly utilizing a pre-segmentation model to carry out preliminary pre-segmentation, screening pre-segmentation regions, then carrying out gradient calculation on the pre-segmentation regions to obtain a main concentrated part of a misdiagnosis region, namely the edges of tissue regions, and calculating the edge regions by utilizing a high-precision fine segmentation model to realize accurate segmentation of the canceration regions. The invention combines the rapidity of the pre-segmentation model and the accuracy of the fine segmentation model, can realize the canceration region segmentation algorithm which can meet the actual application level, can be applied to the actual social production process, effectively reduces the detection workload and accelerates the diagnosis speed.

Riaz Z, Khan B, Abdullah S, Khan S, Islam MS reported on Lung Tumor Image Segmentation from Computer Tomography Images Using MobileNetV2 and Transfer Learning. An improved hybrid neural network via the fusion of two architectures, MobileNetV2 and UNET, for the semantic segmentation of malignant lung tumors from CT images was developed. The transfer learning technique was employed and the pre-trained MobileNetV2 was utilized as an encoder of a conventional UNET model for feature extraction.

Nishio et. al., (2021) reported on an improved hybrid neural network via the fusion of two architectures, MobileNetV2 and UNET, for the semantic segmentation of malignant lung tumors from CT images. The transfer learning technique was employed and the pre-trained MobileNetV2 was utilized as an encoder of a conventional UNET model for feature extraction.

Yang J, Wu B, Li L, Cao P, Zaiane O. reported on MSDS-UNet: a multi-scale deeply supervised 3D U-Net for automatic segmentation of lung tumor in CT. The study presents an effective 3D U-Net equipped with ResNet architecture and a two-pathway deep supervision mechanism to increase the network's capacity for learning richer representations of lung tumors from global and local perspectives.
Despite widespread systems disclosed in the prior arts having their own advantages, there is an emergence for an improved system for lung cancer segmentation and detection, as the above disclosed systems have few limitations.

Drawbacks of the existing systems:

• The existing system is a challenging task for lung cancer tumors to map the location of the tumor regions, shape, and position.
• The difficulty of computation in the segmentation model while training and learning the network.
• Imbalance of data in the training and testing of the model.
• Over fitting the data occurs upon usage of large data during training. Inadequate current data and hidden data can lead to noise and outliers.
• Training the optimal hyper parameters using deep-learning models is challenging in the existing methodology.
Thus, there exists a need in the state of art for an alternative system for lung cancer segmentation with heightened accuracy in detecting lung cancer tumours.

Hence, an attempt has been made to develop a system for lung cancer segmentation using deep learning techniques overcoming above said drawbacks.

OBJECT OF THE INVENTION:

The main object of the present invention is to develop an efficient and highly accurate system for lung cancer segmentation.

Another object of the present invention is to develop a system for lung cancer segmentation using deep learning techniques.
Yet another object of the present invention is to develop a system for lung cancer segmentation employing Parallel Multi-Phase Pivot Point Lung Cancer Network (3D PivotpointNET).

Further object of the present invention is to utilize the developed system to detect lung cancer accurately.

BRIEF DESCRIPTION OF THE DRAWINGS:

Figure 1 depicts basic workflow of the system of the present invention.

Figure 2 depicts detailed workflow of system of the present invention.

SUMMARY OF THE INVENTION:

The present invention discloses a system for lung cancer segmentation using deep learning techniques. The system of the present invention comprising;

a. performing tumor region extraction and tumor shape measurement from lung CT input images using a Pre-processing module;
b. transforming the pre-processed images into a 3D array format adapted by a Reshaping module;
c. training a Parallel Multi-Phase Pivot Point Lung Cancer Network (3D PivotpointNET) model on the reshaped images to generate input feature maps with height, width, and depth;
d. processing the input feature maps using convolutional layers with window sliding sizes of 3x3, 5x5, and 7x7 within a Convolutional processing module;
e. executing tensor dot product operations and concatenating layers for each 3D patch of features through a Patch processing module; and
f. predicting tumor regions accurately by processing output feature maps with global average pooling, wherein a 3D function based on a feature map in every layer is performed to differentiate between lung segmenting tumors and ground truth lung tumors, and classified segmented tumor features with hyper parameter tuning within a Prediction module.

DETAILED DESCRIPTION OF THE INVENTION:

The present invention discloses a system for lung cancer segmentation using deep learning techniques.

The system of the present invention comprising;

a. tumor region extraction and tumor shape measurement from lung CT input images using a Pre-processing module;
b. transforming the pre-processed images into a 3D array format adapted by a Reshaping module;
c. training a Parallel Multi-Phase Pivot Point Lung Cancer Network (3D PivotpointNET) model on the reshaped images to generate input feature maps with height, width, and depth;
d. processing the input feature maps using convolutional layers with window sliding sizes of 3x3, 5x5, and 7x7 within a Convolutional processing module;
e. executing tensor dot product operations and concatenating layers for each 3D patch of features through a Patch processing module; and
f. predicting tumor regions accurately by processing output feature maps with global average pooling, wherein a 3D function based on a feature map in every layer is performed to differentiate between lung segmenting tumors and ground truth lung tumors, and classified segmented tumor features with hyper parameter tuning within a Prediction module.

The pre-processing method of system of the present invention is categorized into two folds: 1. Extraction of tumor regions and measuring the tumor shape. Correlation features measure the tumor texture based on various features for lung CT input images. The tumor shape measures the distance, perimeter, and area. Overlapping tumor regions were calculated by the distance of Bhattacharyya to reshape the tumor, as shown in Figure 1.

After reshaping the 3D array image format trained with 3D PivotpointNET generates an input feature map of height, width, and depth. It has three convolutional layers of window sliding size 3X3, 5X5, and 7X7 over the reshaped input feature map. Every possible location is extracting the 3D patch of features (height, width, and depth). Each 3D patch is performed with a tensor dot product of the kernel, and concatenating the layer for every patch. Every output feature map is processed with global average pooling corresponding to the exact location in the input feature map to predict the tumor accurately, as shown in Figure 2.

Method for segmentation of lung cancer images using deep learning techniques:

a. Pre-processing lung CT input images to extract tumor regions and measure tumor shape;
b. Reshaping the pre-processed images into a 3D array format;
c. Training a Parallel Multi-Phase Pivot Point Lung Cancer Network (3D PivotpointNET) model on the reshaped images to generate input feature maps with height, width, and depth;
d. Applying convolutional layers with window sliding sizes of 3x3, 5x5, and 7x7 over the input feature maps;
e. Performing tensor dot product operations and concatenating layers for each 3D patch of features;
f. Processing output feature maps with global average pooling to predict tumor regions accurately, wherein it is performed as a 3D function based on a feature map in every layer to differentiate between lung segmenting tumors and ground truth lung tumors, and classifies segmented tumor features with hyper-parameter tuning.

The Parallel Multi-Phase Pivot Point Lung Cancer Network model has multiple convolutional layers with varying kernel sizes for feature extraction from the input feature maps. The algorithm contains down-sampling, which is an encoding process of the pivot point to segregate the actual data and synthetic data. The up-sampling is a decoding process of 3D PivotpointNET to determine synthetic data from the output layer. It is performed as a 3D function based on a feature map in every layer to differentiate between lung segmenting tumors and ground truth lung tumors. Thus, it classifies segmented tumor features with hyper parameter tuning. Therefore, the present invention architecture significantly enhances the tumor regions in 3D segmentation performance.

Advantages of the present invention:

• The 3D PivotpointNET architecture of the present invention extracts tumor regions, especially tiny tumors in lung CT images and reshapes the 2D tumour regions into 3D array format in step by step process.

• The 3D PivotpointNET architecture is used to perform large datasets in every encoding and decoding layer with the dot product of kernel to concatenate with a global average of each layer to predict to extract feature maps of the actual data and synthetic data through the decoding process. It can handle the imbalance of data and optimize the hyperparameter while training the model and extracting the tumour accurately.

• 3D pivotpoint method of the present invention can extract the patch-by-patch feature map by segmenting tumor regions very fast.

• The system of the present invention can map the location of the tumor regions, shape, and position easily.

In the pre-processing step of the present invention employs techniques for extracting tumor regions based on correlation features and measuring tumor shape based on distance, perimeter, and area. The Parallel Multi-Phase Pivot Point Lung Cancer Network model has multiple convolutional layers with varying kernel sizes for feature extraction from the input feature maps. The down-sampling the feature maps to encode tumor region. The up-sampling the feature maps to decode tumor regions. The processing output feature maps employ a softmax classification function for predicting tumor regions. Reshaping the pre-processed images into a 3D array format includes identifying diagonal axis reshaped tumor regions to transform into 3D. The pre-processing step encompasses measuring overlapping tumor regions using Bhattacharyya distance. The processing output feature maps executehyperparameter tuning to optimize tumor segmentation accuracy. The convolutional layers employReLU activation functions. Tensor dot product operations performed by applying a kernel to each 3D patch of features. Processing output feature maps encompass segmenting lung nodules.

From the above, it is concluded that the system and method of the present invention be a better alternative for lung cancer segmentation and detection as the algorithm employed in the present invention identifies the diagonal axis reshaped tumor regions processed with input slices which can transform into 3D.

In one of the preferred embodiment, the present invention discloses a system for lung cancer segmentation using deep learning techniques. The system of the present invention comprising;

a. performing tumor region extraction and tumor shape measurement from lung CT input images using a Pre-processing module;
b. transforming the pre-processed images into a 3D array format adapted by a Reshaping module;
c. training a Parallel Multi-Phase Pivot Point Lung Cancer Network (3D PivotpointNET) model on the reshaped images to generate input feature maps with height, width, and depth;
d. processing the input feature maps using convolutional layers with window sliding sizes of 3x3, 5x5, and 7x7 within a Convolutional processing module;
e. executing tensor dot product operations and concatenating layers for each 3D patch of features through a Patch processing module; and
f. predicting tumor regions accurately by processing output feature maps with global average pooling, wherein a 3D function based on a feature map in every layer is performed to differentiate between lung segmenting tumors and ground truth lung tumors, and classified segmented tumor features with hyperparameter tuning within a Prediction module.

As per the invention, in the system of the present invention, the tumor region extraction performed by the pre-processing module employs techniques for extracting tumor regions based on correlation features and measuring tumor shape based on distance, perimeter, and area.

In accordance with the invention, in the system of the present invention, the 3D PivotpointNET model has multiple convolutional layers with varying kernel sizes for feature extraction from the input feature maps.

In accordance with the invention, in the system of the present invention, encoding tumor regions in the feature maps are performed by a down-sampling module.

In accordance with the invention, in the system of the present invention, decoding tumor regions in the feature maps are performed by an up-sampling module.

In accordance with the invention, in the system of the present invention, the prediction module encompasses a softmax classification functions for predicting tumor regions.

In accordance with the invention, in the system of the present invention, the reshaping of pre-processed images into a 3D array format by the Reshaping module encompasses identifying diagonal axis reshaped tumor regions to transform into 3D.

In accordance with the invention, in the system of the present invention, the pre-processing module encompasses techniques for measuring overlapping tumor regions using Bhattacharyya distance

In accordance with the invention, in the system of the present invention, hyperparameter tuning to optimize tumor segmentation accuracy is executed within the Prediction module.

In accordance with the invention, in the system of the present invention, the convolutional layers in the convolutional processing module employs ReLU activation functions.

In accordance with the invention, in the system of the present invention, a kernel is applied to each 3D patch of features during tensor dot product operations within the Patch processing module.

In accordance with the invention, in the system of the present invention, segmenting of lung nodules is performed by the prediction module.

Although the invention has now been described in terms of certain preferred embodiments and exemplified with respect thereto, one skilled in the art can readily appreciate that various modifications, changes, omissions, and substitutions may be made without departing from the scope of the following claims.
, Claims:WE CLAIM:

1. A system for lung cancer segmentation using deep learning techniques, comprising:

a. performing tumor region extraction and tumor shape measurement from lung CT input images using a Pre-processing module;
b. transforming the pre-processed images into a 3D array format adapted by a Reshaping module;
c. training a Parallel Multi-Phase Pivot Point Lung Cancer Network (3D PivotpointNET) model on the reshaped images to generate input feature maps with height, width, and depth;
d. processing the input feature maps using convolutional layers with window sliding sizes of 3x3, 5x5, and 7x7 within a Convolutional processing module;
e. executing tensor dot product operations and concatenating layers for each 3D patch of features through a Patch processing module; and
f. predicting tumor regions accurately by processing output feature maps with global average pooling, wherein a 3D function based on a feature map in every layer is performed to differentiate between lung segmenting tumors and ground truth lung tumors, and classified segmented tumor features with hyperparameter tuning within a Prediction module.

2. The system as claimed in Claim 1, wherein the tumor region extraction performed by the pre-processing module employs techniques for extracting tumor regions based on correlation features and measuring tumor shape based on distance, perimeter, and area.
3. The system as claimed in Claim 1, wherein the 3D PivotpointNET model has multiple convolutional layers with varying kernel sizes for feature extraction from the input feature maps.

4. The system as claimed in Claim 1, wherein encoding tumor regions in the feature maps are performed by a down-sampling module.

5. The system as claimed in Claim 1, wherein decoding tumor regions in the feature maps are performed by an up-sampling module.

6. The system as claimed in Claim 1, wherein the said prediction module encompasses a softmax classification functions for predicting tumor regions.

7. The system as claimed in Claim 1, wherein the said reshaping of pre-processed images into a 3D array format by the Reshaping module encompasses identifying diagonal axis reshaped tumor regions to transform into 3D.

8. The system as claimed in Claim 1, wherein the said pre-processing module encompasses techniques for measuring overlapping tumor regions using Bhattacharyya distance.

9. The system as claimed in Claim 1, wherein hyperparameter tuning to optimize tumor segmentation accuracy is executed within the Prediction module.

10. The system as claimed in Claim 1, wherein the said convolutional layers in the convolutional processing module employReLU activation functions.

11. The system as claimed in Claim 1, wherein a kernel is applied to each 3D patch of features during tensor dot product operations within the Patch processing module.
12. The system as claimed in Claim 1, wherein segmenting of lung nodules is performed by the prediction module.

Dated this 02nd day of SEP 2024

For UNIVERSITY OF MADRAS
By its Patent Agent

Dr.B.Deepa
IN/PA 1477