A novel liver tumor classification using improved probabilistic neural networks with Bayesian optimization

Abstract

One of the most common forms of cancer is liver cancer, and successful treatment depends greatly on an early diagnosis. The two primary steps in liver tumor segmentation are liver identification and then tumor segmentation. A sophisticated meta-heuristic technique called Improved Probabilistic Neural Networks and Bayesian Optimization (IPNN-BO) is also introduced to carry out the best classification. In the study, liver computed tomography (CT) scan pictures were gathered for training purposes and added to the LiTs17 dataset. Utilizing performance matrices like accuracy, precision, recall, specificity, sensitivity, and F Measure, the suggested technique is implemented in MATLAB 2021(a). Modern technologies such as KNN, CNN, and DCNN, among others, are used to organize and validate possible policies. The comparison results reveal that the proposed technique outperforms and outperforms the most recent methodologies. IPNN-BO may obtain validation accuracy of up to 99.25 % after applying Bayesian optimization, while KNN, CNN, and DCNN only manage validation accuracy of 88 %, 92.85 %, and 93.75 %, respectively.

Keywords

• Liver tumor
• CT images
• ResUNet model
• Probabilistic neural network and Bayesian optimization