Hyderabad Researchers Develop AI Model to Detect Breast Cancer From Mammograms

The researchers developed the model, named Fuzzy RS-Net, to identify potential signs of breast cancer in mammograms.

Researchers from Hyderabad have developed an artificial intelligence (AI)-based model to detect breast cancer from mammogram images, achieving 94.9% accuracy while requiring less processing time and memory than several other models evaluated in the study.

The study, titled “Fuzzy-based residual shufflenet-based breast cancer detection using mammogram images,” was conducted by Kumari Jelli and Pavan Kumar Pagadala from the Department of Computer Science and Engineering at Koneru Lakshmaiah Education Foundation. It was published in Scientific Reports, part of Nature Portfolio.

The researchers developed the model, named Fuzzy RS-Net, to identify potential signs of breast cancer in mammograms. The system incorporates a method for handling uncertainty in medical images, where abnormalities may not always be clearly distinguishable.

Model Achieves 94.9% Accuracy

The model was evaluated using the Curated Breast Imaging Subset of the Digital Database for Screening Mammography, a publicly available mammogram dataset. It recorded 94.9% accuracy, 95.8% sensitivity, and 93.8% specificity.

Sensitivity measures the model's ability to correctly identify cancer cases, while specificity measures how accurately it identifies cases without cancer.

The researchers also evaluated Fuzzy RS-Net using other publicly available mammography datasets and across different datasets. According to the study, the model required less processing time and memory compared with several AI models used for benchmarking.

The system first removes unwanted noise from mammogram images, identifies areas of potential concern, and then analyzes image patterns in those regions to classify the mammograms. An evaluation of the model's individual components showed that each contributed to its overall performance. Statistical testing also indicated significant improvements over the comparison models, with p-values below 0.05.

Researchers Call for Hospital Validation

Despite the results, the researchers said the model has so far been tested only on publicly available datasets. They recommended validation using larger and more diverse patient datasets to address potential dataset bias and assess whether the findings can be reproduced in real-world settings.

The researchers also recommended testing the system in hospitals before clinical use and incorporating explainable AI tools to help clinicians understand how the model reaches its findings.

The study cited World Health Organization data showing that more than 2.3 million women were diagnosed with breast cancer in 2020, while more than 685,000 deaths were reported that year. The researchers noted that analyzing large volumes of mammograms manually can be time-consuming, while subtle abnormalities can be difficult to identify.

Stay tuned for more such updates on Digital Health News