“Review and Analysis of CNN Approach for Lung Cancer Detection and Classification Using Deep Learning”

Abstract

Lung cancer is one of the most frequent malignancies, with 225,000 diagnoses, 150,000 deaths, and a $12 billion annual health-care expenditure. In this paper, we provide a computer-aided diagnostic (CAD) technique for classifying CT scans with unmarked nodules for lung cancer classification. To distinguish lung tissue from the rest of the CT data, thresholding was used as an initial segmentation strategy. The next best lung segmentation came via Thresholding. The original plan was to send segmented CT scans directly into 3D CNNs for categorization, but this proved insufficient. Instead, prospective nodule candidates in CT images were identified using an updated U-Net trained on LUNA16 data (CT scans with labelled nodules). Because the U-Net nodule detection produced a number of false positives when determining whether a CT scan was positive or negative for lung cancer, regions of CTs with segmented lungs where the most likely nodule candidates were located as defined by the U-Net production were fed into 3D Convolutional Neural Networks.

Networks of Neurons (CNNs). 3D CNNs were used to generate the Accuracy O Test Set. Our CAD system surpasses prior literature CAD systems by containing only three crucial phases (segmentation, nodule candidate identification, and malignancy classification), allowing for more effective training and detection, as well as more generalisation for diverse cancers.