Intervention of Deep Learning Model for Predicting Lung Cancer Using Histopathological Images

Abstract

Lung cancer remains one of the leading causes of cancer-related deaths worldwide, with accurate and timely diagnosis being crucial for effective treatment. This thesis presents a novel approach to lung cancer subtype classification using histopathological images and deep learning techniques. Chapter 1 introduce the significance of lung cancer detection and the potential of automated classification systems in assisting pathologists. The chapter outlines the research objectives and the potential impact of this study on improving diagnostic accuracy and efficiency. Chapter 2 provides a comprehensive literature review, examining existing methods for lung cancer detection and classification. It highlights the advancements in deep learning techniques, particularly the use of convolutional neural networks in medical image analysis. Chapter 3 details the methodology employed in this study. We describe the modified U-Net architecture used for lung cancer subtype classification and explain four different preprocessing techniques: no pre-processing, Super-Resolution Convolutional Neural Network (SRCNN), Super-Resolution Generative Adversarial Network (SRGAN), and Contrast Limited Adaptive Histogram Equalization (CLAHE). In Chapter 4, we present and analyze the experimental results. The performance of our model is evaluated using metrics such as accuracy, precision, recall, F1 score, specificity, and AUCROC score. We compare the effectiveness of different pre-processing techniques and provide visual representations of our results through accuracy/loss curves, ROC curves, and confusion matrices. Research findings indicate that the SRGAN pre-processing method yielded the highest overall accuracy of 96% in classifying lung cancer subtypes. The model demonstrated robust performance across all three subtypes, with particular strength in distinguishing between subtypes Lung Benign and Adenocarcinomas. Chapter 5 concludes the thesis by summarizing the key findings and discussing their implications for lung cancer diagnosis. We also address the propose directions for future research.

This study contributes to the field of medical image analysis by demonstrating the potential of deep learning models in lung cancer subtype classification. The proposed method shows promise in assisting pathologists, potentially leading to more accurate and efficient diagnoses.