A bio-inspired deep learning methodology aimed at optimizing predictive outcomes for liver cancer by improving segmentation and classification processes. It features a hybrid segmentation algorithm, SegNet- UNet-ABC, for extracting liver lesions from computed tomography (CT) images. Initially, the SegNet network is used to segment the liver from CT scans, followed by the UNet network, which focuses on detecting and outlining lesions within the liver. A major innovation in this work is the hybridization of the Artificial Bee Colony (ABC) optimization algorithm with both the SegNet and UNet networks, allowing fine- tuning of their hyper parameters to enhance segmentation accuracy. As hyper parameter selection plays a critical role in performance, this optimization strategy is crucial for achieving better results. The deep learning model developed has been shown to outperform existing methods in terms of automatic detection rates, classification accuracy, and execution speed for liver cancer detection and classification. It also employs various classification techniques, including Fisher’s Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and Multilayer Perceptron (MLP), with training algorithms such as Levenberg-Marquardt (MLP-LM) and Bayesian Regularization (MLP-RBF). The approach also integrates Otsu’s global thresholding technique, morphological operations, and watershed transformation, all of which help ensure precise identification of liver lesions. The results show that the proposed system significantly outperforms traditional methods, offering a more reliable, faster, and more accurate solution for liver cancer detection.
Keyword: SegNet, UNet, DNN, Artificial Bee Colony(ABC) Optimization, Segmentation.