AI-Driven Analysis of Digital Pathology Images to Improve Pediatric Sarcoma Subtyping

Pediatric sarcomas are rare and diverse tumors that can develop in various types of soft tissue, such as muscle, tendons, fat, blood or lymphatic vessels, nerves, or the tissue surrounding joints. These sarcomas are classified into different subtypes based on several factors, including the tissue of origin and various molecular characteristics. Accurately classifying a patient’s sarcoma subtype is crucial as it helps guide treatment decisions and optimize outcomes. Unfortunately, due to the heterogeneity of sarcomas, classification can be extremely challenging. This often requires complex molecular and genetic testing, along with external review by highly specialized pathologists who rely on pattern recognition skills developed over years of training. Such resources are not always available in many healthcare settings. Now, findings of a study presented at the American Association for Cancer Research (AACR) Annual Meeting 2025 have shown that an artificial intelligence (AI)-based model can accurately classify pediatric sarcomas using only digital pathology images.

In their study, researchers at UConn Health (Farmington, CT, USA) and their collaborators explored the potential of AI to identify pediatric sarcoma subtypes with high precision. They used 691 digital images of pathology slides from various collaborators, representing nine distinct sarcoma subtypes, to train AI algorithms to detect patterns specific to each subtype. By digitizing tissue pathology slides, the researchers were able to convert the visual information a pathologist typically examines into numerical data that a computer can process. Much like how smartphones can identify a person’s face in photos and organize them into albums, the AI models learned to recognize tumor morphology patterns in the digitized slides and categorize them into diagnostic groups linked to specific sarcoma subtypes. To ensure consistency, the researchers developed and applied open-source software to harmonize images collected from different institutions, accounting for differences in format, staining, magnification, and other variables. The harmonized images were then broken into small tiles, which were analyzed using deep learning models that extracted numerical data for further evaluation using a novel statistical method.

This statistical method generated feature summaries for each slide, which were then assessed by the trained AI algorithms to assign the slides to specific subtypes. In validation experiments, the AI models successfully identified sarcoma subtypes with high accuracy. Specifically, the models distinguished between Ewing sarcoma and other sarcoma types in 92.2% of cases, non-rhabdomyosarcoma soft tissue sarcomas and rhabdomyosarcoma soft tissue sarcomas in 93.8% of cases, alveolar rhabdomyosarcoma and embryonal rhabdomyosarcoma in 95.1% of cases, and alveolar rhabdomyosarcoma, embryonal rhabdomyosarcoma, and spindle cell rhabdomyosarcoma in 87.3% of cases. A limitation of the study was the relatively small number of available pathology images for training the AI algorithms. However, the researchers pointed out that, given the rarity of pediatric sarcomas, their imaging dataset is likely the largest multicenter collection of pediatric sarcomas to date, encompassing a wide range of subtypes, anatomical locations, and patient demographics.

“Our findings demonstrate that AI-based models can accurately diagnose various subtypes of pediatric sarcoma using only routine pathology images. This AI-driven model could help provide more pediatric patients access to quick, streamlined, and highly accurate cancer diagnoses regardless of their geographic location or health care setting,” said Adam Thiesen, an MD/PhD candidate at UConn Health. “Our models are built in such a way that new images can be added and trained with minimal computational equipment,” he added. “After the standard data processing, clinicians could theoretically use our models on their own laptops, which could vastly increase accessibility even in under-resourced settings.”

Related Links:
UConn Health