Researchers led by Ahmad P. Tafti, assistant professor, Department of Health Information Management, and director of the Pitt HexAI Lab, and Johannes F. Plate, associate professor, Department of Orthopaedic Surgery, and HexAI Lab associate director, in collaboration with colleagues across engineering, orthopedics and clinical practice, recently published a study in npj Health Systems introducing KneeXNet-2.5D, an explainable artificial intelligence (AI) system designed to automatically analyze and segment knee MRI scans.
The work aims to help clinicians more efficiently and consistently identify damage to knee cartilage and meniscus, the key structures involved in osteoarthritis and sports injuries. Traditionally, outlining these structures on MRI requires time-intensive manual work by domain experts and radiologists. KneeXNet-2.5D developed by scientists at the University of Pittsburgh uses explainable AI algorithms to perform this task in seconds, while also highlighting areas where the model is uncertain, allowing clinicians to better understand and trust the results.
This project is a strong example of AI-driven innovation grounded in real clinical needs. Rather than focusing on complex “opaque box” models, the team designed an efficient and explainable AI strategies that works well even with low computing resources, an important consideration for real-world health care settings, including community and rural hospitals. The system not only produces accurate results but also provides visual cues that help clinicians and surgeons interpret how and why the AI reached its conclusions, supporting safer and more transparent use of AI in medicine.
Equally important, this work reflects cross-disciplinary team science in action. The project brought together expertise from AI and computer vision, health informatics, medical imaging, computer science, orthopedic surgery, physical therapy and clinical research. Junior faculty and undergraduate and graduate students at SHRS played central roles in developing and validating the AI models, curating and annotating MRIs, conducting rigorous evaluations and translating technical advances into clinically meaningful AI-powered tool sets. Their leadership, supported by senior faculty mentorship including James Irrgang, Department of Physical Therapy professor and senior advisor, and clinical partners highlights SHRS’ commitment to training the next generation of researchers and scientists at the intersection of AI, health systems and patient-centered care.
The KneeXNet-2.5D, is publicly and freely available for any academic, research and educational purposes.
For more information about Tafti’s work and his team, visit the PittHexAI Research Laboratory website.
