A research team spearheaded by Harvard Medical School (HMS) in the US has created an AI tool, dubbed Pathology Image Characterization Tool with Uncertainty-aware Rapid Evaluations (PICTURE), to differentiate brain cancer types.
The tool is capable of differentiating between glioblastoma and primary central nervous system lymphoma (PCNSL), notably two brain cancers that are often confused due to their similar appearance under the microscope.
Go deeper with GlobalData
Discover B2B Marketing That Performs
Combine business intelligence and editorial excellence to reach engaged professionals across 36 leading media platforms.
Glioblastoma is known as a widespread aggressive brain tumour, originating from brain cells, while PCNSL, a less common type, is said to arise from immune cells.
HMS said that misdiagnosis between these two can lead to inappropriate treatment strategies, which can have severe implications for patients.
The National Institutes of Health has partly funded this work.
HMS’ PICTURE tool also features an uncertainty component that flags unfamiliar tumour types, prompting a review by medical professionals.
US Tariffs are shifting - will you react or anticipate?
Don’t let policy changes catch you off guard. Stay proactive with real-time data and expert analysis.
By GlobalDataThe institute said during brain surgery, tumour tissues are often rapidly assessed by freezing them in liquid nitrogen and examining them under a microscope – a process that can alter cellular features but offers a swift assessment within approximately 15 minutes.
This initial diagnosis can change after a more thorough examination by pathologists in the following days. The PICTURE model aims to reduce the uncertainty and risk of error at this crucial stage.
Tested across five hospitals, PICTURE is said to have “outperformed both human pathologists and other AI models”.
HMS Blavatnik Institute’s biomedical informatics associate professor and study senior author Kun-Hsing Yu said: “Our model can minimise errors in diagnosis by distinguishing between tumours with overlapping features and help clinicians determine the best course of treatment based on a tumour’s true identity.”
HMS added that the model, co-developed by Yu and co-first authors Shih-Yen Lin and Junhan Zhao, was trained and assessed on 2,141 brain pathology slides from around the world, including rare cases and both frozen and formalin-fixed samples.
While the current focus is on glioblastoma and PCNSL, there are plans to expand the tool’s capabilities to other cancer types and integrate genetic and molecular data for more comprehensive analysis.
The team said that the AI model is publicly accessible, allowing other scientists to utilise and enhance it further.
