![JMA – Ninja & Deep Jandu [Official MV] Kaptaan](https://i.ytimg.com/vi/-dsfIh319s0/maxresdefault.jpg "JMA – Ninja & Deep Jandu [Official MV] Kaptaan")
Hyderabad: Researchers in Hyderabad have developed an artificial intelligence-based system to detect breast cancer from mammogram images, reporting nearly 95% accuracy while using less computing power and memory than several other models tested in the study.
The study, “Fuzzy based residual shufflenet based breast cancer detection using mammogram images”, by Kumari Jelli and Pavan Kumar Pagadala of the Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, was published in Scientific Reports of Nature Portfolio.
The researchers developed a model called Fuzzy RS-Net to help identify signs of breast cancer in mammograms. The system is designed to improve image analysis while reducing the computing resources required. It also uses a method to deal with uncertainty in mammogram images, where abnormalities may be difficult to distinguish clearly.
Nearly 95% accuracy
When tested on the Curated Breast Imaging Subset of the Digital Database for Screening Mammography, a publicly available mammogram database, the model recorded 94.9% accuracy, 95.8% sensitivity and 93.8% specificity.
In simple terms, sensitivity measures how well the system identifies cancer cases, while specificity indicates how well it correctly recognises cases without cancer.
The researchers also tested the model on other public mammography datasets and across different datasets. They reported that it required less processing time and memory than several other artificial intelligence models used for comparison.
The system first reduces unwanted noise in mammogram images, identifies areas that may require attention and then analyses patterns in those areas to classify the images. The researchers also tested the contribution of different parts of the system and found that each helped improve its overall performance.
Statistical tests showed that the improvement over the comparison models was significant, with p-values below 0.05. The model also maintained its performance when tested on different datasets.
Hospital testing still needed
The authors, however, said the system has so far been tested only on publicly available databases. They recommended further validation using larger and more diverse sets of patient data to reduce possible dataset bias and determine whether the results can be repeated in real-world conditions.
They also called for testing the system in hospitals before its clinical use and for adding explainable artificial intelligence tools so doctors can better understand how the model arrives at its findings.
The study cites World Health Organization figures showing that more than 2.3 million women were diagnosed with breast cancer in 2020, with over 685,000 deaths that year. The researchers said early detection is important, while examining large numbers of mammograms manually can be time-consuming and subtle abnormalities may be difficult to identify.
