Computer-aided diagnosis for early detection and staging of human pancreatic tumors using an optimized 3D CNN on computed tomography

It is challenging to screen for and treat pancreatic cancer (PC), an extremely malignant tumor, both early in the course of the disease and as part of the later stages of treatment. In this article, a computer-aided diagnosis (CAD) technique for detecting PC early, particularly when assessing PC staging, is suggested. With this, clinical staff will be better equipped to give a treatment plan and can intervene with therapy early on as a result. This article addresses the process of effectively segmenting and classifying a pancreatic tumor using a deep learning (DL) network by following the four stages outlined below. Initially, computed tomography (CT) images are used for diagnosis, which is obtained from TCIA public access. Following the raw image acquisition, these images need to be pre-processed using the Boosted Anisotropic Diffusion Filter (BADF) and Contrast Limited Adaptive Histogram Equalization (CLAHE). Then, using the DMFCM segmentation approach, the images are segmented. Through Bag of Visual Words (BOVW) and Support Vector Machine (SVM), features are extracted, and the best features are given as input to the classifier. Eventually, classification is carried out using optimized 3D convoluted neural networks (3D CNN) using Improved Harris Hawks Optimization (IHHO). The implemented model achieved better results of 98.32% accuracy, 99% sensitivity, and 99% specificity. The proposed model is compared to some cutting-edge models such as normal CNNs, LSTMs, RESNETs, RNNs, and 3D CNNs to determine which one performs well. In terms of specificity, sensitivity, accuracy, and recall, the proposed model scored better than other models.