|How accurate is the AI at classifying normal breast MRI scans?|
How well does the AI classify normal breast MRI exams?
An Artificial Intelligence (AI) algorithm could classify 20% of breast MRI exams as normal without missing any cancer, New York City researchers say. If used in clinical practice, AI could make breast MRI screening more efficient.
In a presentation at the International Society for Magnetic Resonance Imaging in Medicine (ISMRM) meeting in London, researchers at Memorial Sloan Kettering Cancer Center in New York City discussed how they developed a deep learning model designed to classify MRI scans. normal breast MRI to a special degree. worklist that only requires an abbreviated review by a radiologist.
In tests, the algorithm worked well and would have generated an estimated 20% time savings for radiologists, according to presenter Arka Bhowmik, PhD.
“We developed a deep learning tool that classifies the 20% of normal cases on an abbreviated worklist without missing any cancers in the test set,” he said.
Breast MRI is used to screen for breast cancer in high-risk patients, but more than 98% of these exams are normal. In the study presented at ISMRM 2022, researchers sought to use AI to rank completely normal exams on an abbreviated radiologist worklist; they also compared the algorithm’s performance to that of fellowship-trained radiologists. They also wanted to estimate the projected time savings from using AI.
The researchers developed their deep learning model based on the use of the BI-RADS system to provide AI training tags. BI-RADS 1 cases were considered to have no imaging findings, while the remaining BI-RADS scores (2-6) were considered to have imaging findings.
The researchers retrospectively collected 16,020 contrast-enhanced breast axial MRI exams performed at their institution on 8,330 patients between 2013 and 2019. Of these, 12,911 (80%) exams were used for training, 1,627 (10%) were used for validation and 1482 (10%) were reserved for testing. In addition, 50 exams were randomly selected from the test pool for a study of readers.
The test set included 1,467 exams without cancer and 15 with cancer. The AI algorithm detected all 15 cancer cases and classified the 20% of non-cancer exams into an abbreviated interpretation worklist for a radiologist. The remaining 80% of the exams were graded for full interpretation by a radiologist.
The researchers calculated that the total projected reading time under this paradigm would have been reduced from 148 hours to 119 hours, a time savings of 20%.
In a study of readers comparing radiologists’ performance with the AI algorithm, 42 of 50 MRI exams were free of cancer and eight included cancer. The eight cancer exams were identified by both the radiologists and the AI algorithm.
Of the 42 cancer-free exams, nine were classified according to an abbreviated worklist for radiologists and 33 were classified according to the complete radiologist’s interpretation. Radiologists ruled out 39 (92%) of the 42 cancer-free exams and marked three (7%) for biopsy.
In the next phase of their work, the researchers are now performing a multi-institutional validation of the algorithm. They are also developing a more generalized end-to-end algorithm that includes an initial breast segmentation step, according to the authors.
Source: AuntMinnie.com, Direct News 99