Radiologists help develop AI tool that detects patients in danger of domestic violence
Radiologists have helped to develop an artificial intelligence tool that helps identify patients in danger of experiencing domestic abuse.
Experts with Mass General Brigham, Boston, recently detailed their work on Friday in NPJ Women’s Health. Such “intimate partner violence” is a widespread but underreported public health concern, with about one-third of women experiencing it sometime in their life.
Early detection can be critical to preventing worsening health among those impacted. But doing so can be hard when victims hesitate to disclose abusive relationships, due to safety concerns or financial dependency.
Mass General experts developed an AI program to address this issue, using inputs including radiology reports, imaging and diagnoses to aid in detection.
“With access to the imaging history of patients, radiologists have an advantage in recognizing signs of [intimate partner violence],” the study’s authors wrote. “Patterns in radiology studies—including high frequency of radiology imaging and injuries to the face, neck and upper extremities—can indicate the likelihood of IPV. However, time constraints, silos of subspecialization, cognitive overload and the urgency to address immediate symptoms often limit the ability of radiologists and clinicians to leverage this information.”
Emergency radiologist and senior author Bharti Khurana, MD, MBA, and colleagues sought to address this by developing an automated decision support tool that aids in prediction. Khurana and colleagues at MIT trained three different machine-learning models for their working, using electronic health records from 673 women who visited a domestic abuse center between 2017 and 2022. They also added further data from 4,169 demographically matched controls who did not report intimate partner violence. AI programs tested included a “tabular” model using structured EHR data including diagnoses, medications and social deprivation index based on ZIP code. A second AI model used unstructured radiology and ED reports, while a third “fusion” model combined both types.
All three models demonstrated high accuracy when validated via a separate dataset. However, the hybrid model achieved the highest accuracy at approximately 88%. When tested on time-stamped, archived medical records, the fusion model predicted nearly 81% of cases sooner, at an average of over 3.7 years before a patient sought medical care. Researchers also further validated the AI model with data from additional patient groups and discovered “similarly high accuracies.”
“Our research offers proof of concept that AI can support clinicians in flagging possible abuse earlier," senior author Khurana said in a statement March 13. “Earlier identification of intimate partner violence and future risk may enable clinicians to intervene sooner and help prevent significant mental and physical health consequences.”
Khurana also has previously published research on this topic, including a 2025 JACR study detailing signs providers can spot that may signal intimate partner violence. This latest work noted that those with mental health disorders, chronic pain and frequent ED visits were most likely to experience IPV. Conversely, those who regularly accessed screening mammography and other preventive services were less likely.
“It is also important to emphasize that the models are not intended to provide a definitive diagnosis of IPV, but rather to serve as a tool to assist healthcare providers in having patient-centered conversations around IPV,” the authors added.
Marty Stempniak has covered healthcare since 2012, with his byline appearing in the American Hospital Association's member magazine, Modern Healthcare and McKnight's. Prior to that, he wrote about village government and local business for his hometown newspaper in Oak Park, Illinois. He won a Peter Lisagor and Gold EXCEL awards in 2017 for his coverage of the opioid epidemic.