Artificial Intelligence Tool Expands Identification of Steatotic Liver Disease in Radiology Reports

Recent findings from a real-world analysis indicate that an artificial intelligence platform can significantly increase the identification of patients with metabolic dysfunction-associated steatotic liver disease by scanning existing ultrasound radiology reports. The evaluation, conducted by health-tech company Briya, demonstrated a 360% expansion in the relevant study population when utilizing natural language processing technology to extract hidden diagnostic data.

Uncovering Hidden Data in Unstructured Reports

The analysis examined 28,795 abdominal ultrasound reports encompassing 20,422 unique patients, gathered between 2020 and 2024. Utilizing the company AIRE platform, researchers identified an additional 4,036 patients with a steatotic liver, expanding the study population from an initial 1,122 up to 5,158 individuals. According to the study lead Gadi Lalazar, who serves as the head of the liver unit at Israel Shaare Zedek Medical Center, many of these individuals were previously unaware of their condition because the relevant findings were embedded within ultrasound reports generated for entirely unrelated medical examinations.

Researchers noted that early identification creates a critical window for physicians to initiate interventions before the liver condition progresses toward more severe and potentially irreversible stages. However, the analysis also highlighted a substantial gap when relying solely on structured electronic healthcare record systems. Specifically, only 22% of the additionally identified patients had a corresponding diagnosis documented within structured electronic healthcare records. The remaining 78% of patients were surfaced exclusively through the natural language processing analysis of unstructured text data.

Evaluating Risk and Future Implications

Unstructured reports typically consist of traditional imaging summaries dictated by radiologists in a narrative format, contrasting with rigid structured summary templates. Following the expansion of the study cohort, researchers utilized existing laboratory data and ultrasound reports to assess disease progression risk. The assessment revealed 4,368 patients with a Fibrosis-4 Index score placing them above conventional at-risk thresholds for liver scarring.

• Briya researchers utilized specialized agents to extract information from unstructured fields.
• Data was harmonized with existing structured records to construct the expanded study cohort.
• Investigators aim to present the complete findings at an upcoming medical meeting.

Study authors emphasized that unlocking insights from legacy radiology reports allows investigators to build larger and more representative patient populations without necessitating additional data collection phases. Dr Or Shaked, director of medical research solutions at Briya, noted that automating complex data preparation tasks enables researchers to dedicate more effort toward addressing clinical questions that may ultimately influence patient care. Broader market trends compiled by GlobalData indicate continued expansion within the healthcare artificial intelligence sector, though clinical adoption requires ongoing validation of real-world retrospective findings.

Source: Medical Device Network