AI-Based Model Accurately Classifies Pediatric Sarcomas Using Digital Pathology Images

An artificial intelligence (AI)-based model successfully classified pediatric sarcomas using only digital pathology images, according to findings presented at the American Association for Cancer Research (AACR) Annual Meeting. The study was led by Dr. Jill Rubinstein, a surgical oncologist and senior research scientist at The Jackson Laboratory, and used software developed by Dr. Sergii Domanskyi, associate computational scientist at the same institution. The research received support from the National Institutes of Health, The Jackson Laboratory, and Hartford Hospital.

AI Model Accurately Classified Pediatric Sarcoma Subtypes with up to 95.1 Percent Precision in Validation Studies

In this study, Adam Thiesen, an MD/PhD candidate at UConn Health and The Jackson Laboratory working in the lab of Jeffrey Chuang, PhD, and colleagues investigated the use of artificial intelligence (AI) to classify pediatric sarcoma subtypes. Using 691 digital pathology slide images provided by Massachusetts General Hospital, Yale New Haven Children’s Hospital, St. Jude Children’s Research Hospital, and the Children’s Oncology Group, the team trained AI algorithms to recognize patterns associated with nine distinct sarcoma subtypes.

Thiesen explained that by digitizing tissue pathology slides, the researchers converted visual data typically analyzed by pathologists into numerical data that could be processed by AI. He likened the approach to how smartphones identify faces in photos. The AI model recognized tumor morphology patterns in the digitized images and grouped them into diagnostic categories linked to specific sarcoma subtypes.

To ensure consistency across data sources, the team developed and used open-source software to harmonize images from different institutions, accounting for differences in format, staining, and magnification. They then divided the harmonized images into smaller tiles and processed them through deep learning models that extracted numerical features. A novel statistical method summarized these features, which trained AI algorithms used to classify each slide by sarcoma subtype.

In validation experiments, the AI algorithms demonstrated high accuracy in classifying pediatric sarcoma subtypes, according to Thiesen. The models correctly distinguished Ewing sarcoma from other sarcoma types in 92.2% of cases, non-rhabdomyosarcoma soft tissue sarcomas from rhabdomyosarcoma subtypes in 93.8% of cases, alveolar rhabdomyosarcoma from embryonal rhabdomyosarcoma in 95.1% of cases, and among alveolar, embryonal, and spindle cell rhabdomyosarcoma in 87.3% of cases.

Largest Multicenter Dataset of Pediatric Sarcoma Images Supports Future Development of AI-Assisted Pathology Tools

Thiesen reported that AI-based models developed in the study accurately diagnosed several pediatric sarcoma subtypes using only routine digital pathology images. The researchers highlighted the potential of this approach to deliver faster and highly accurate cancer diagnoses without requiring advanced imaging tools or specialized infrastructure.

The model was designed to be lightweight and adaptable. According to Thiesen, clinicians could integrate new pathology images with minimal computational equipment. After standard data processing, the model could theoretically run on a personal laptop, expanding access to AI-supported diagnostics in under-resourced or remote clinical settings. 

One limitation of the study was the relatively small number of pathology images available for training the algorithms. However, Thiesen pointed out that pediatric sarcomas are rare, and the research team’s dataset may be the largest multicenter imaging collection of its kind to date. It included a diverse range of tumor subtypes, anatomical locations, and patient demographics, offering a strong foundation for further development and validation of AI-assisted pathology tools. “We hope that, over time, additional groups will work with us to further increase the size of this dataset,” said Thiesen.

Related posts:

Senti Biosciences to Showcase New Cell & Gene Therapy Data at AACR Annual Meeting 2025 Study Shows ctDNA Liquid Biopsy Detects Colorectal Cancer Recurrence Earlier Than Imaging Immunai and AstraZeneca’s $18 Million AI Collaboration to Transform Cancer Drug Development BPIPO and DCB Collaborate with NVIDIA to Host AI Forum, Focusing on Cross-Domain Integration for Precision Health

Author

Denisse Sandoval