Enhancing breast cancer diagnosis using deep learning and gradient multi-verse optimizer: a robust biomedical data...

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Breast cancer (BC) is a leading cause of mortality in women, necessitating improved early detection methods.
• Computer-aided diagnosis (CAD) systems, particularly those using deep learning (DL), show promise for accurate BC diagnosis.
• Current DL training optimizers (gradient-based and metaheuristic) often stagnate in local optima, limiting model performance.

Purpose of the Study:

• To propose a hybrid optimization algorithm combining gradient search with the multi-verse optimizer (MVO) for DL models.
• To develop a deep neural network (DNN) for BC diagnosis optimized by the novel hybrid algorithm.
• To evaluate the proposed optimizer's effectiveness in improving DL model accuracy and convergence speed for BC detection.

Main Methods:

• A hybrid optimizer integrating gradient search into the multi-verse optimizer (MVO) was developed.
• A three-hidden-layer deep neural network (DNN) was trained using the proposed hybrid optimizer for BC diagnosis.
• Experimental validation was performed on the Wisconsin Breast Cancer Dataset (WBCD) and Wisconsin Diagnosis Breast Cancer (WDBC) datasets.

Main Results:

• The proposed hybrid optimizer demonstrated superior performance over traditional gradient-based and metaheuristic methods.
• The DL model optimized by the hybrid approach achieved high accuracy (93.5% on WBCD, 96.73% on WDBC) with rapid convergence (6 epochs).
• Exceptional performance metrics were recorded, including high sensitivity, specificity, precision, F1 score, and MCC on both datasets.

Conclusions:

• The proposed hybrid optimization algorithm effectively addresses local optima stagnation in DL training for high-dimensional spaces.
• This research significantly advances CAD systems for breast cancer, offering a more accurate and efficient diagnostic tool.
• The developed optimizer shows substantial potential for enhancing medical classification tasks in diverse healthcare applications.