Segmentation and Classification of Lung Nodule Using Categorical Hinge Loss Function-based Convolutional Neural Network.

The accurate segmentation and classification of lung nodules using Computed Tomography (CT) scans play a significance part for early diagnosis of lung cancers. But, there are risks of misclassification that occurs due to incorrect suspiciousness of malignancy, alongside the reliance on the nodule size. In this research, the Categorical Hinge Loss Function based Convolutional Neural Network (CHLF-CNN) is proposed for multi-classification of lung nodule. This research comprises five main stages: Primarily, data collection is carried out through the Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) as well as Lung Nodule Analysis 2016 (LUNA16) of CT-scan image for validating the model's effectiveness. Next, pre-processing is performed by deploying Median Filter based image denoising and Min-max normalization technique. Then, the pre-processed data is segmented by employing the Firefly Algorithm (FA) with the Cuckoo Search Algorithm (CSA) based Markov Random Field (MRF) approach. Then, feature extraction is carried out by utilizing the shape and texture-based features. Finally, classification is performed to classify the lung nodules into multi-classes. The CHLF-CNN attains a commendable accuracy of 96.54% and 99.56% respectively on the LIDC-IDRI and LUNA16 dataset, as opposed to the existing methods, Deep CNN and Morphology-based CNN (2PMorphCNN). [ABSTRACT FROM AUTHOR]