AI Tool May Improve Tumor Removal Accuracy During Breast Cancer Surgery

Researchers from the University of North Carolina (UNC) Department of Surgery, the Joint UNC-North Carolina State University (NCSU) Department of Biomedical Engineering, and the UNC Lineberger Comprehensive Cancer Center have developed an artificial intelligence (AI) tool to predict whether cancer tissue has been fully removed during breast cancer surgery.

The model is designed to help detect the presence of cancer cells that may not be visible to the naked eye.

“Some cancers you can feel and see, but we can’t see microscopic cancer cells that may be present at the edge of the tissue removed. Other cancers are completely microscopic,” explained Kristalyn Gallagher, DO, section chief of breast surgery in the Division of Surgical Oncology and UNC Lineberger member, who served as the study’s senior author. “This AI tool would allow us to more accurately analyze tumors removed surgically in real-time, and increase the chance that all of the cancer cells are removed during the surgery. This would prevent the need to bring patients back for a second or third surgery.”

Typically, breast cancer surgery involves resecting the tumor and taking a small amount of surrounding tissue—known as the surgical margin—to help ensure that the cancerous tissue has been fully removed. The tumor specimen is then photographed with a mammography machine.

The resulting imaging is then reviewed by the care team to make sure the abnormal tissue has been removed. Then, the mammography results are further analyzed by pathologists, who work to gauge whether the cancer cells extend beyond the tumor’s edge, or pathological margin.

If cancer cells do extend beyond the pathological margin, additional breast cancer tissue may still remain in the patient.

The problem with this process is that it can take up to a week to complete, meaning that a patient may need to undergo another surgery to remove the remaining cancer cells.

The AI model addresses this by leveraging specimen mammography to predict the pathologic margin status of resected breast tumors almost instantly. Specimen mammography involves using an X-ray to photograph the tumor specimen and can be performed immediately in an operating room, meaning that the AI could be deployed to assess tumor margins during surgery.

The model was trained, validated, and tested using a dataset of specimen mammography images matched with pathologic margin status collected at UNC from 2017 to 2020. Patient data, including demographics and tumor size and shape were also incorporated.

When tested, the tool demonstrated a sensitivity of 84 percent, a specificity of 42 percent, and an area under the receiver operating characteristic curve of 0.71.

The researchers hope that the model may provide high-quality clinical decision support.

“It is interesting to think about how AI models can support doctor’s and surgeon’s decision making in the operating room using computer vision,” said first author Kevin Chen, MD, general surgery resident in the Department of Surgery. “We found that the AI model matched or slightly surpassed humans in identifying positive margins.”

The tool may be particularly useful in clinical settings with fewer resources.

“It is like putting an extra layer of support in hospitals that maybe wouldn’t have that expertise readily available,” said Shawn Gomez, EngScD, professor of biomedical engineering and pharmacology and co-senior author on the paper. “Instead of having to make a best guess, surgeons could have the support of a model trained on hundreds or thousands of images and get immediate feedback on their surgery to make a more informed decision.”

Moving forward, the research team plans to continue feeding the model data to improve its performance, but the tool will need to be further validated before it can be deployed clinically.