CNN-BO-LSTM: an ensemble framework for prognosis of liver cancer

Abstract

Computer-assisted diagnosis (CAD) is preferred for cancer identification across the globe, which relies on computerized image processing. The creation of previous CAD instruments involved a semi-automated approach that employed traditional deep learning techniques. Such techniques are not well versed with CAD instruments in terms of accuracy. Therefore, the given manuscript presents a convolutional neural network normalized architecture embedded with Bayesian optimization and long short-term memory (CNN-BO-LSTM) for the identification of liver cancer. Early pre-processing is done on the input magnetic resonance imaging (MRI) scan images to improve clarity. Next, we used dynamic binary classification to apply the accurate and region of interest (ROI) extracting approach. It is followed by automatic retrieval of CNN-based appearances from the ROI approach. For classification purposes, LSTM is used, which categorizes the images as benign or malignant. The proposed design’s testing outcomes, which combine characteristics with CNN-based ROI extraction and LSTM classification, surpassed the current state-of-the-art techniques.