Researchers prove early detection of breast cancer is possible using AI

Posted Today

Researchers at the City College of New York and Memorial Sloan Kettering Cancer Center (MSK) are raising hopes that AI can improve early detection of breast cancer. 

Researchers at CCNY & Sloan Kettering have proven that early detection of breast cancer is possible using AI.

Researchers have developed an AI algorithm and evaluated its ability to identify breast cancer on MRI scans one year in advance.  

The AI model was trained on MRI from 52,598 breasts and fine-turned on 3,209 scans from 910 patients at high risk of developing cancer. The data contained 115 cancers that were diagnosed one year after a normal MRI result.

According to the study published in Academic Radiology, the algorithm can detect cancers one year earlier than current clinical practice. It was used to rank the top 10% of the highest-risk MRIs. If those MRIs had been analyzed by a radiologist, early detection could have been increased by up to 30%.

The AI was also able to identify the region where cancer would be detected in 66 of the 115 cases. A radiologist identified the cancers in 83 of the 115 cases. The radiologist and AI agreed on 54 cases.

“These findings are one of the first large scale demonstrations of the possibility of implementing AI tools for early detection of cancer in high-risk women from an MRI screening program,” said Lukas Hirsch, a postdoctoral researcher in CCNY’s Parra Lab with a PhD in biomedical engineering, who is a lead author of the study.

“Working with MRI is difficult in general due to the difficulty in obtaining a sufficiently large number of images to train these large AI models, so research and development of tools for this is just starting,” he added. “We have shown that AI is very good at detecting early stages of cancer but there are still more efforts to do in improving performance and  implementing this in clinics.”

While the data acquisition, storage and annotation for the project is taking place at MSK, the methods and AI-tools are developed at CCNY.

Source: CCNY