An artificial intelligence (AI) model developed by researchers at Cedars-Sinai can now automate cancer staging using patients' pathology reports, potentially accelerating the selection process for clinical trials. This innovation, published in Nature Communications, addresses the slow and tedious nature of traditional cancer staging methods that rely on manual review of patient records.
The AI model, named BB-TEN (Big Bird—TNM staging Extracted from Notes), extracts and interprets text from pathology reports to quickly identify the cancer stage. This eliminates the need for specially trained personnel to manually sift through laboratory reports and clinicians' notes, a process that can take months.
AI Model Development and Validation
The AI model is based on a transformer architecture, designed to mimic human decision-making. It was initially trained using publicly available pathology reports from The Cancer Genome Atlas, encompassing nearly 7,000 patients and 23 cancer types. Subsequent validation on nearly 8,000 pathology reports from a single medical center demonstrated high accuracy, as measured by standard AI evaluation metrics.
"This was an important finding because it means that our AI model is an 'off- the-shelf' tool that can be generalized to other institutions without requiring that it be trained for each location," said Nicholas Tatonetti, Ph.D., vice chair of Computational Biomedicine at Cedars-Sinai.
Potential Applications and Impact
Beyond clinical trial screening, the AI model can automate patient classification for observational and retrospective data analysis, as well as potential treatments. The researchers suggest that future work could integrate pathology text with other clinical data to further personalize cancer treatment.
Jason Moore, Ph.D., chair of the Department of Computational Biomedicine at Cedars-Sinai, emphasized the potential of the AI model to speed up the selection of candidates for cancer clinical trials, accelerating the development of relevant treatments and making them available to more patients. The AI model is available to other institutions for academic uses.
The creation of the AI model was made possible by earlier research that removed technical obstacles to computers extracting and analyzing pathologists' notes from electronic health records.
