Computer Program Combines AI and Heat-Imaging Technology for Early Breast Cancer Detection

In 2020, an estimated 2.1 million new cases and 627,000 deaths were reported by the World Health Organization (WHO), highlighting a rising incidence in many low- and middle-income countries. While mammography is a highly effective tool for early detection of breast cancer, its accessibility is limited due to cost and availability constraints. Now, researchers have developed a tool powered by machine learning that could serve as a complementary, non-invasive, and pain-free alternative to mammography for early breast cancer detection.

A group of researchers, led by Nanyang Technological University (NTU, Singapore), created the computer program to identify potential tumors in the human breast. This innovation is based on the understanding that malignant breast tumors distribute heat differently compared to healthy breast tissue. The program, named Physics-informed Neural Network (PINN), integrates artificial intelligence (AI) with heat-imaging technology. Developed in collaboration with medical doctors specializing in breast imaging and intervention, PINN analyzes thermal infrared images of the breast, detecting heat patterns to identify possible malignant tumors within five minutes. To refine and 'train' PINN, the team fed it with infrared breast scans of thousands of patients, both with and without malignant breast tumors.

Upon testing PINN on hundreds of infrared breast images that contained malignant tumors, the researchers discovered that the program could accurately identify harmful tumors with 91% accuracy. Unlike traditional methods, PINN does not require bulky equipment and operates much faster, using an infrared camera to capture images of the breast from multiple angles for computer analysis. Since it employs heat-imaging technology, it presents a safer alternative for women at higher risk of breast cancer or with a family history of the disease, especially considering that mammograms involve exposure to ionizing radiation. However, the researchers emphasize that PINN is not intended to replace current diagnostic techniques. Instead, it can act as a valuable and accessible tool for the early detection of breast cancer.

“Our study’s findings, and the development of PINN, centers around AI’s ability to swiftly and accurately analyze vast datasets, specifically thousands of infrared breast scans,” said Associate Professor Eddie Ng Yin Kwee, from the School of Mechanical and Aerospace Engineering at NTU Singapore, who led the study. “We also benefited from machine learning when calibrating PINN as it made the program easily trainable, aiding it to recognize patterns and generalize well to new, unseen data, making it adaptable and reliable. PINN could assist in the early identification of potential abnormalities in breast tissues, not only contributing to better treatment outcomes but also streamlines the screening process, allowing healthcare professionals to prioritize complex cases.”

Related Links:
NTU Singapore