Mammogram mass segmentation and detection using Legendre neural network-based optimal threshold.

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Medical Imaging Analysis

Background:

• Breast cancer is a significant global health concern, with early detection crucial for reducing mortality rates.
• Machine learning models offer advanced capabilities for analyzing complex biomedical data, including mammograms.
• Accurate segmentation and detection of mammographic masses are vital for effective radiologist-assisted diagnosis.

Purpose of the Study:

• To develop an optimal adaptive thresholding method for segmenting and detecting masses in mammograms.
• To enhance the accuracy of breast cancer diagnosis through improved image analysis.
• To assist radiologists by providing a more reliable tool for mammogram interpretation.

Main Methods:

• A single-layer Legendre neural network was employed for model development.
• The Block Based Normalized Sign-Sign Least Mean Square (BBNSSLMS) algorithm was utilized for supervised training.
• The Legendre neural network expands input vectors using Legendre polynomials with a recursive weight update principle.

Main Results:

• The proposed model achieved a sensitivity of 95% and an accuracy of 96% on a dataset of 151 mammograms from the MIAS database.
• The method demonstrated reduced computational complexity for threshold selection.
• False positives per image were calculated at 1.19, indicating effective mass detection.

Conclusions:

• The developed Legendre neural network model with BBNSSLMS algorithm provides an effective approach for mammogram mass segmentation and detection.
• The optimal adaptive thresholding method enhances diagnostic accuracy and aids radiologists in early breast cancer detection.
• The sparse nature of the adaptive model contributes to faster convergence and computational efficiency.