AI ups breast cancer detection post-mastectomy
Artificial intelligence as standalone reader may improve the detection of second breast cancers in women who have undergone mastectomy surgery.
Even with thorough treatment and surgery, women with a personal history of breast cancer (PHBC) are at increased risk of developing second cancers in the future. These patients are recommended to undergo more frequent surveillance mammograms following surgery to ensure there are no signs that cancer is developing.
While this method is effective, it presents challenges for radiologists and, historically, interpretations of unilateral mammograms post-mastectomy have yielded lower sensitivity. A new paper in the journal Radiology suggests that this is an area where AI-enabled reads could offer improvements.
“These cancers are associated with poor prognostic markers, including larger tumor size, negative hormone receptor status, and positive lymph node status. Hence, there is a need for more precise and effective tools for screening and surveillance mammography to improve patient outcomes,” Jung Min Chang, with the department of radiology at Seoul National University Hospital in the Republic of Korea, and colleagues explain. “Given the growing population of patients with a PHBC undergoing surveillance mammography, investigating how this unique group can benefit from AI is highly important.”
The team sought to compare the performance of a standalone algorithm to that of human radiologists. To do this, they retrospectively applied the tool to the exams of more than 4,000 women who had undergone post-mastectomy surveillance mammography between January 2011 and March 2023.
The AI’s sensitivity and specificity were compared to the original reporting radiologists’ interpretations, while outcomes were determined using follow-up data in the patients’ electronic health records.
Just under 3% of the patients were diagnosed with a second breast cancer. AI detected 17.4 per 1,000 exams, yielding a sensitivity of 65.8%, while radiologists identified 14.6 second cancers per 1,000 mammograms, at a sensitivity of 55%, respectively. Though AI outperformed rads in sensitivity, the humans were able to perform slightly better in terms of specificity.
The algorithm detected 32% of cancers missed by radiologists, though both the AI and radiologists missed a little more than 30% of the cancers that were later detected and determined to be present in the exams included in the analysis.
“Our findings provide support for the use of AI in mammographic surveillance for patients treated with unilateral mastectomy, revealing that AI substantially increases cancer detection,” the group notes.
The authors acknowledge that their study was limited by its retrospective nature, but are encouraged by their findings, nonetheless. Their work highlights some areas where AI can be further fine-tuned to improve its specificity in this specific cohort of patients and they are hopeful that prospective studies will confirm their findings.
The study abstract is available here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.