Enhancing Breast Cancer Data Classification Performance in Medical Data by Employing GRU-LMA Deep Learning Classifier for Accurate Diagnosis

Abstract

Objectives: To suggest a new data classifier to enhance breast cancer data classification in mammography digital images (MDI) in order to reduce the error rate and training time and improve accuracy in breast cancer prediction. A deep learning approach is utilized to identify the significant deep spots in MDI for accurate diagnosis. Methods: The Gated Recurrent Unit (GRU)-based DL approach is employed to speed up the learning phase in the DRNN classifier, and the Levenberg-Marquardt algorithm (LMA) is used to identify and reduce errors during classification. For feature extraction, noise removal, & filtering, Adaptive Median Filtering (AMF) and Tetrolet transform (TTM) algorithms are utilized, and the features are separated into two categories (morphology & texture). The King Abdulaziz University Breast Cancer Mammogram Dataset (KAUBCMD) dataset is used for this research work. KAUBCMD has 1416 case reports, with 2 views for both the right and left breasts, and 5662 MDI with clinical results of 3 target classes (negative, benign, and malignant) and 2 subclasses (incomplete and suspected malignant). MATLAB tool is used to evaluate the DL-based GRU-LMA approach based on performance metrics. The detailed comparative analysis is done with prevailing classifier models such as linear SVM, KNN-HP, and K-Means GRI. Findings: The suggested DL-based GRU-LMA classifier model shows proven results with 95.19% accuracy, 96.03% TPR and 95.09% TNR, 95.07% sensitivity, 97.81% specificity, 96% precision, 97.03% recall, 17 minutes of training time, 3% error rate, 3% FPR, and 4% FNR. GRU-LMA outperforms the current classifier models L-SVM, KNN-HP, and K-Means GRI. Novelty: The proven results show breast cancer prediction and classification accuracy with minimal time and error rate, which helps the radiologist and other BC clinical experts make diagnoses easily. The GRU-LMA addresses the limitations of the prevailing classifier models, L-SVM, KNN-HP, and K-Means GRI.https://www.indjst.org/