Krill herd optimization algorithm with deep convolutional neural network fostered breast cancer classification using mammogram images

• Published in: Concurrency and computation Vol. 35; no. 7
• Main Authors: P, Pratheep Kumar, V, Mary Amala Bai, Krish, Ram P.
• Format: Journal Article
• Language: English
• Published: Hoboken, USA John Wiley & Sons, Inc 25.03.2023; Wiley Subscription Services, Inc
• Subjects: Algorithms; Artificial neural networks; Breast cancer; Compressing; Datasets; deep convolutional neural network classifier; Dynamic range; Image classification; Image compression; Image enhancement; Krill; krill herd optimization algorithm; Mammography; Neural networks; Optimization; Optimization algorithms; Real time; real time dataset
• ISSN: 1532-0626; 1532-0634
• DOI: 10.1002/cpe.7605

Summary In this paper proposes a Krill Herd Optimization algorithm with Deep Convolutional neural network fostered Breast Cancer Classification using Mammogram Images (BC‐APPDRC‐DCNN‐KHO). Here, the input images are taken from Real time and MAMMOSET datasets. These images are pre‐processed using Altered Phase Preserving Dynamic Range Compression (APPDRC) technique. This APPDRC is applied for preserving local features, compressing dynamic range of images, and enhancing the speckle noise filtering, these are all necessary for better boundary detection. Then, the Pre‐processed images are classified using Deep Convolutional neural network (DCNN). The DCNN weight parameters are optimized based on Krill Herd Optimization algorithm. The Proposed BCC‐DCNN‐KHO‐MI method classifies the input breast cancer imageries into 3 categories: benign, malignant, and normal. The proposed BCC‐DCNN‐KHO‐MI method in Real time dataset attains 18.505%, 19.45%, 16.19%, 17.56% and 16.19% higher accuracy; 15.38%, 12.06%, 12.71%, 26.62% and 18.902% higher Precision; 3.12%, 10.52%, 13.57%, 22.75% and 14.93% higher F‐score, 59.56%, 41.25%, 56.47%, 42.36% and 37.27% lower computation time; 23.87%, 21.87%, 32.87%, 42.76% and 21.05% higher AUC compared with the existing methods, like BCC‐Google Net‐MI, BCC‐Visual Geometry Group Network‐MI, BCC‐Residual Networks‐MI, BC‐RERNN‐LOA‐MI and BC‐CNN‐MI respectively.