Using artificial intelligence, study achieves nearly 99% accuracy in classifying mammograms

17 Jun 2026

Breast cancer is the most common cancer among women worldwide, including in the Philippines, and early detection is crucial in saving lives. While mammograms are the standard tool for screening, they are often difficult to interpret and can lead to missed cases or false alarms. This research explores how artificial intelligence (AI) can support radiologists by improving the accuracy and reliability of mammogram analysis.

The study used eight advanced computer models, known as Convolutional Neural Networks (CNNs), which are designed to recognize patterns in images. Each CNN was trained to extract important features from mammogram images. These features were then combined, or “concatenated,” to capture more sample comprehensive information than any single model could provide. To refine the process, two feature selection methods were applied: Mutual Information, which measures how much a feature contributes to classification, and the Best-Worst Method, a decision-making approach that prioritizes the most informative features.

The results showed that the system could classify mammograms with nearly 99% accuracy, significantly higher than what individual models achieved. The unique aspect of this study is its integration of multiple CNNs with two advanced feature selection strategies – an approach rarely applied in breast cancer imaging. While Mutual Information produced the highest raw accuracy, the Best-Worst Method proved more efficient by achieving strong results with fewer features.

By combining accuracy with efficiency, this research demonstrates a novel way to design AI-powered diagnostic tools. Such tools could help hospitals, especially in resource-limited settings, improve early detection and ultimately save more lives.

This research contributes to the advancement of AI in healthcare by improving the accuracy, efficiency, and interpretability of breast cancer diagnosis from mammograms. By combining multiple convolutional neural networks (CNNs), applying feature concatenation, and introducing a comparative feature selection framework (Best-Worst Multi-Attribute Decision-Making vs. Mutual Information), the study demonstrates that diagnostic models can achieve near-perfect classification accuracy (up to 99%) while also reducing computational complexity.

The significance lies in three areas:

Clinical Impact – The framework offers potential for earlier and more reliable non-invasive breast cancer detection, which can directly improve patient survival rates, reduce the need for non-directed invasive procedures, and lower healthcare costs.

Methodological Innovation – This is among the few studies that integrate multi-CNN feature fusion with advanced feature selection methods, offering a scalable and efficient diagnostic tool for high-dimensional medical imaging data.

National and Global Relevance – Breast cancer remains the most diagnosed cancer among women worldwide, including the Philippines. The proposed approach provides a foundation for AI-driven diagnostic support systems that can augment radiologists, especially in resource-constrained settings.

Overall, the study demonstrates how carefully designed AI models can significantly enhance diagnostic accuracy, efficiency, and trustworthiness, thereby contributing to both the scientific community and the healthcare sector.

Authors: Eusib Vincent J. Pulvera and Demelo M. Lao (Department of Computer Science, College of Science, University of the Philippines Cebu)

Read the full paper: https://doi.org/10.1109/ICIIBMS62405.2024.10792816

Image by Klaus Nielsen from Pexels

Using artificial intelligence, study achieves nearly 99% accuracy in classifying mammograms

Breast cancer is the most common cancer among women worldwide, including in the Philippines, and early detection is crucial in saving lives. While mammograms are the standard tool for screening, they are often difficult to interpret and can lead to missed cases or false alarms. This research explores how artificial intelligence (AI) can support radiologists by improving the accuracy and reliability of mammogram analysis.

The study used eight advanced computer models, known as Convolutional Neural Networks (CNNs), which are designed to recognize patterns in images. Each CNN was trained to extract important features from mammogram images. These features were then combined, or “concatenated,” to capture more sample comprehensive information than any single model could provide. To refine the process, two feature selection methods were applied: Mutual Information, which measures how much a feature contributes to classification, and the Best-Worst Method, a decision-making approach that prioritizes the most informative features.

The results showed that the system could classify mammograms with nearly 99% accuracy, significantly higher than what individual models achieved. The unique aspect of this study is its integration of multiple CNNs with two advanced feature selection strategies – an approach rarely applied in breast cancer imaging. While Mutual Information produced the highest raw accuracy, the Best-Worst Method proved more efficient by achieving strong results with fewer features.

By combining accuracy with efficiency, this research demonstrates a novel way to design AI-powered diagnostic tools. Such tools could help hospitals, especially in resource-limited settings, improve early detection and ultimately save more lives.

This research contributes to the advancement of AI in healthcare by improving the accuracy, efficiency, and interpretability of breast cancer diagnosis from mammograms. By combining multiple convolutional neural networks (CNNs), applying feature concatenation, and introducing a comparative feature selection framework (Best-Worst Multi-Attribute Decision-Making vs. Mutual Information), the study demonstrates that diagnostic models can achieve near-perfect classification accuracy (up to 99%) while also reducing computational complexity.

The significance lies in three areas:

Clinical Impact – The framework offers potential for earlier and more reliable non-invasive breast cancer detection, which can directly improve patient survival rates, reduce the need for non-directed invasive procedures, and lower healthcare costs.

Methodological Innovation – This is among the few studies that integrate multi-CNN feature fusion with advanced feature selection methods, offering a scalable and efficient diagnostic tool for high-dimensional medical imaging data.

National and Global Relevance – Breast cancer remains the most diagnosed cancer among women worldwide, including the Philippines. The proposed approach provides a foundation for AI-driven diagnostic support systems that can augment radiologists, especially in resource-constrained settings.

Overall, the study demonstrates how carefully designed AI models can significantly enhance diagnostic accuracy, efficiency, and trustworthiness, thereby contributing to both the scientific community and the healthcare sector.

Authors: Eusib Vincent J. Pulvera and Demelo M. Lao (Department of Computer Science, College of Science, University of the Philippines Cebu)

Read the full paper: https://doi.org/10.1109/ICIIBMS62405.2024.10792816

Image by Klaus Nielsen from Pexels