Enhancing early breast cancer diagnosis through automated microcalcification detection using an optimized ensemble...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast cancer diagnosis relies on accurate detection of microcalcifications.
• Deep learning models show promise but face challenges due to architectural diversity.
• An ensemble approach is proposed to enhance precision and dependability.

Purpose of the Study:

• To develop an optimized deep learning ensemble model for mammogram preprocessing and microcalcification detection.
• To improve the accuracy and reliability of breast cancer diagnostics.

Main Methods:

• A preprocessing framework involving artifact removal, segmentation, and filtering was implemented.
• Transfer learning with ResNet-50 and ensemble optimization of AlexNet, GoogLeNet, VGG16, and ResNet-50 were utilized.
• Hyperparameter optimization and rigorous evaluation protocols were employed.

Main Results:

• The ensemble model achieved an average confidence score of 0.9305 for microcalcification classification.
• Normal cases showed an average confidence of 0.8859, indicating consistent predictions.
• High performance metrics including accuracy, precision, recall, F1-score, and AUC were attained.

Conclusions:

• The ensemble model significantly enhances the accuracy and dependability of breast cancer diagnostics.
• Focus on average confidence ratings improves clinical diagnosis effectiveness.
• This novel methodology offers substantial improvements in microcalcification detection.