Weight Optimized Neural Network Using Metaheuristics for the Classification of Large Cell Carcinoma and Adenocarcinoma from Lung Imaging

Background: Neural Networks are utilized in several applications in the field of healthcare, one such being the classification of lung cancers. Innovative advancements in diagnosing tumours are a major boost to developing novel treatment techniques in the early stages of lung cancer. Method: In this work, a novel image-based features selection method for classifying lung Computed Tomography (CT) images is introduced. A new fusion-based technique through a combination of Gabor filters and first order histograms was created. The suggested model utilizes Multi-Layer Perceptron Neural Networks (MLP-NN) alongside Krill Herds (KH) for structural optimization which consists of three phases. First, images are pre-processed and features selected through the new fusion-based selection method. Next, the selected features are got through the application of Correlation based Feature Selection (CFS), Mutual Information (MI), Fuzzy Unordered Rule Induction Algorithm (FURIA) that choose the highest ranked ones. Lastly, classifiers like AdaBboost or MLP-NN carry out the classification of the cancers. Results: Misclassification rates, average true positive rates, average false discovery rates are utilized in the evaluation of CFS-Furia, CFS-AdaBoost classifier, CFS-MLPNN, CFS-KHMLPNN, MI-Furia, MIAdaBoost classifier, MI-MLPNN and MI-KHMLPNN. The suggested MI-KHMLPNN outperforms all others in every category. Conclusion: The model was evaluated with several lung CT images and has proven to attain excellent results in the classification of lung cancers.