A team of researchers at the University of Chicago has developed an artificial intelligence (AI) tool that can help pathologists more accurately diagnose a rare group of cancers known as thymic epithelial tumors (TETs). The study detailing this work was recently published in the Annals of Oncology.
TETs are cancers arising from the thymus, a small gland in the upper chest that plays a critical role in the immune system. In the United States, TETs affect only 2-3 people per million each year. Diagnosing these tumors is challenging due to their rarity and the significant variation in how they appear and behave across patients. Accurate classification into one of five main TET subtypes is essential, as misdiagnosis can affect treatment decisions and patient outcomes.
“This is a very rare type of cancer, so very few people in the world are trained to diagnose and treat it,” said Marina Garassino, MD, Professor of Medicine at UChicago Medicine and senior author of the study. She added that in non-academic settings, diagnostic discrepancies for TETs can be as high as 40%.
To address this challenge, the team developed an AI-based computational model using digital pathology. The model was trained on microscope images of tumors from 119 TET patients sourced from The Cancer Genome Atlas Program (TCGA), with subtypes confirmed by expert pathologists. Testing on 112 additional cases from the University of Chicago demonstrated that the AI tool could classify TET subtypes with high accuracy, particularly excelling at identifying thymic carcinomas, the most aggressive subtype.
“Essentially, we created a tool that, in the hands of a non-expert pathologist, can correctly diagnose 100% of thymic carcinomas and outperform non-expert diagnoses,” said Garassino.
Importantly, the AI tool is freely available and designed to support, not replace, pathologists. Its goal is to aid clinicians who do not specialize in rare thymic tumors in making accurate diagnoses and guiding appropriate treatment.
“This project was a true multidisciplinary effort, bringing together data scientists, pathologists, and oncologists to learn from each other about what could—and couldn’t—be achieved,” Garassino said.
The team is now working to validate the tool on a larger scale, incorporating data from additional cancer centers in the U.S. and Europe. One ongoing challenge is harmonizing differences in laboratory and imaging procedures across hospitals, which can affect how tumors appear in digital images. Future iterations of the algorithm aim to correct for these variations, making the tool more widely applicable.
The study, “Deep learning discriminates thymic epithelial tumors’ histological subtypes using digital pathology,” was supported by grants from the National Institutes of Health and a scholarship from Associazione TUTOR (Pierluigi Galli and Eurovetro Recycling SRL). The research was also made possible with support from the Department of Medicine, Section of Hematology/Oncology, and the Department of Pathology at the University of Chicago, as well as the TCGA Research Network.
Additional authors include Matteo Sacco, Anna Di Lello, Alexander McGeough, Alessandra Esposito, Rishi Sharma, Aliya Husain, Qudsia Arif, Maha Elsebaie, Alexander Pearson (UChicago); James Dolezal (Geisinger Cancer Institute); Erica Pietroluongo (University of Naples Federico II); and Mirella Marino (IRCCS Regina Elena National Cancer Institute).
