Deep Features Based Automated Multimodel System for Classification of Non-Small Cell Lung Cancer

Abstract

Machine Learning (ML) and Deep Learning (DL) are the most promising techniques and have a wide variety of applications but in the medical domain, they encounter several challenges including unbalance data, shortage of data, feature redundancy that lead to low accuracy rates. To get complete benefits of these techniques and overcome these challenges, this research investigated various fusion of deep learning-based feature extractors and machine learning-based classifiers for differentiating two subtypes Adenocarcinoma (ADC) and Squamous cell carcinoma (SCC) of non-small cell lung cancer (NSCLC). In the current study, 400 training, 190 validation, and 38 testing computed tomography (CT) scan images of ADC and SCC were used. Five deep feature extractors: InceptionResNetV2, InceptionV3, Xception, VGG16 and, VGG19 and three classifiers: Support vector machine (SVM), XGBoost, and Fully connected neural network (FCNN) were used to generate seven automated integrated models. Then the most effective optimal model was selected via comparative study and performance was further evaluated by accuracy, error rate, sensitivity, precision, specificity, and f1-score. The most effective optimal model InceptionResNetV2 + SVM got training, validation and testing accuracy 93.50%, 93.16%, and 94.74% respectively. Furthermore, on test data sensitivity for ADC was 0.9565 and for SCC was 0.9333 respectively. In the medical domain, the achievements of the DL and ML integrated model deliver an optimal, reliable, convenient, automated system that will help domain experts to make precise decisions in time.