Segmentation and Classification of 3D Lung Tumor Diagnoses Using Convolutional Neural Networks

This work uses data from the 2022 Kaggle Data Science Bowl to provide a CAD system for determining the classification of Computed Tomography (CT) images with unidentified nodules in the context of lung cancer. As a first stage in the segmentation procedure, threshold is used to isolate the tissue in the lungs from the other parts of the CT image. The second-best result for lung segmentation was achieved by thresholding. As a result, the first strategy of feeding the segmented CT images into 3 Dimension Convolution Neural Networks (3D CNNs) for classifications failed. Instead, a customized U-Net trained on LUNA16 data (CT images with tagged nodules) was used to identify potential nodule locations in the Kaggle database CT scans. Due to the high number of false positives generated by the U-Net nodule identification, the output from segmentation CT scans of the lung was fed into lung cancer scan positivity is often determined using 3D convolutional neural networks. In existing method is not compatible to detect lung nodules and does not predict cancer types. A proposed work is Lung Cancer Diagnosis utilizing Heterogeneous Neural Network (LCD-HNN) is a unique approach proposed in this study for early and precise detection. The CT characteristics are extracted using deep neural networks. Detecting malignant cells at an early stage is crucial to saving the patient's life, and this detection relies heavily on the precision of GLCM feature extraction. This research also makes use of a state-of-the-art using a 3D convolutional neural network to boost diagnostic precision. In terms of 3D CNN precision, the test set was 98.6%. Proposed CAD approach outperforms existing CAD systems in the literature since it consists of just three main steps (segmentation, nodule candidate identification, and malignancy classification) and so requires less labeling data for training and detection and may be utilized for a broader range of malignancies. Differentiating between both benign and malignant tumors is also possible using the proposed method. Standard statistical methods are used to analyze the data, and the findings show that the suggested hybrid DL methodology is effective at detecting lung cancer at an early stage.