Your Donations in Action: Feyisope R. Eweje, BS
Deep learning model for MR-based classification of bone lesions
![Feyisope R. Eweje, BS](https://www.rsna.org/-/media/images/rsna/news-articles/2024/august/eweje-feyisope.gif?rev=a0a00d14a1854760abe7c8ec98c207b1&la=en&h=190&w=135&hash=189BBA842B2B905221820D27C14BE30D)
![R&E Foundation 40th anniversary logo](https://www.rsna.org/-/media/images/rsna/news-articles/2024/march/re-40th-anniversary-logo.gif?rev=feab1d9b271641699daaa97b4aef1fe1&la=en&h=190&w=135&hash=85343F5026FC72D33B17275AC0156D23)
Bone tumors are a leading cause of cancer-related deaths in people under the age of 20 in the U.S. They are difficult to classify on imaging because of tumor heterogeneity, tumor rarity and similarity in appearance of malignant tumors to non-cancerous conditions, such as bone infections.
For bone lesions that cannot be completely characterized by radiographic imaging, MRI can be useful in providing additional visualization. However, existing evidence demonstrates that while MRI has high sensitivity for characterizing malignant bone tumors, it has low specificity. This leads to a high rate of false positive assessments, creating undue patient anxiety and unnecessary subsequent workup.
For his 2020 RSNA Research Medical Student Grant project, Feyisope R. Eweje, BS, a medical student at the University of Pennsylvania in Philadelphia, and colleagues aimed to develop a novel deep neural network model that would integrate features from multi-sequence 3D MR data and patient clinical data to differentiate benign bone lesions from those that are malignant.
The team developed and tested a multimodal model that can assess the malignant potential of suspicious bone lesions using clinical data from over a thousand cases collected from five hospitals.
“Our findings indicated that the model performs on par with radiologists in terms of accuracy and sensitivity while demonstrating improved specificity,” Eweje said. The results of the study were published in EBioMedicine. If developed for clinical deployment, the AI model could reduce the rate of unnecessary biopsies, improving the patient experience and care delivery. The R&E Foundation grant provided Eweje with essential funding to explore a practical application of AI technology in medical imaging.
“This grant supported my first research work studying the application of AI in imaging diagnostics, laying the groundwork for future efforts investigating AI applications in health care,” Eweje said.
For More Information
Learn more about R&E Foundation funding opportunities.
Read our previous Your Donations in Action story.