Members do research with faculty at Pitt, UPMC, and other schools, in AI and well beyond it. Here's who's working on what. Join a group, or start your own.
Graph neural networks that predict which drug targets could push a diseased cell back toward health.
Isuru Herath
Most models predict how a drug will change a cell. PDGrapher, a causally inspired graph neural network, solves the inverse problem: given a diseased cell's state, it predicts the combination of therapeutic targets that would push it back toward healthy. It found effective targets more often than competing methods, and trains much faster.
Read the paper2025 · Nature Biomedical Engineering
PDGrapher, a causally inspired graph neural network, predicts combinations of therapeutic targets that reverse disease phenotypes, and trains much faster than existing methods.
Outcomes research in neurosurgery, from trigeminal neuralgia to brain tumors, and reviews of neurosurgical care across Africa.
Peace has co-authored a series of reviews that pull together studies from across Africa on spina bifida, spine trauma, brain aneurysms, and craniosynostosis, to show how care and outcomes differ from place to place.
2026 · Clinical Neurology and Neurosurgery
Pain in more than one branch of the trigeminal nerve didn't predict worse outcomes after microvascular decompression than pain in a single branch.
2026 · Journal of Clinical Neuroscience
Reviews immunotherapy studies in high-grade glioma; newly diagnosed IDH-mutant tumors were linked to longer survival than recurrent ones.
2026 · Journal of Clinical Neuroscience
Pools African studies of spina bifida to describe how it presents, how it's managed, and outcomes, which varied by region.
2025 · Journal of Neurosurgery: Pediatrics
Pooled evidence suggests laser thermal therapy shrinks most pediatric brain tumors treated with it; the authors call for prospective trials.
2025 · World Neurosurgery
Compares two nerve-pain drugs around lumbar spine surgery; neither clearly lowered next-day pain, though gabapentin reduced opioid use.
2025 · Neuro-Chirurgie
Reviews published cases of a rare spinal vascular syndrome; surgical treatment was linked to improvement.
2025 · Journal of Clinical Neuroscience
Reviews how ruptured brain aneurysms present and are treated across African studies, where surgical clipping was most common.
2025 · World Neurosurgery
Reviews craniosynostosis care across African countries; nearly all reported patients had surgery, with few complications.
2024 · Neuro-Oncology Advances
Reviews African studies of brain and spinal metastases, which mostly came from breast and lung cancer.
2024 · Journal of Neurosurgery: Spine
Across African studies, spine injuries most often came from road traffic accidents, often with long delays to care.
Cytopathology and surgical pathology, what chatbots like ChatGPT and Llama can and can't do in the field, and reviews that explain AI to clinicians.
Josh PantanowitzJosh re-ran a published Delphi study on the future of pathology, with ChatGPT and Llama in place of the expert panel. The models reached consensus more often than the humans did. It's an early proof of concept that AI might one day help run expert surveys like these.
Read the paper2026 · Journal of Pathology Informatics
Re-runs a published pathology Delphi study with ChatGPT and Llama in place of the expert panel; the models reached consensus more often than the humans did.
2026 · Acta Cytologica
Cancer found in the fluid around the heart mostly spread from lung and breast tumors, and larger fluid samples caught it more often.
2025 · International Journal of Surgical Pathology
Some neuroendocrine tumors unexpectedly stain for a white-blood-cell marker used to spot lymphoma, so pathologists should use broader stain panels.
2025 · Modern Pathology
Reviews LLMs, GANs, and diffusion models in medicine: where they help, the tools available, and concerns about privacy, bias, and cost.
2025 · Modern Pathology
The opening review and glossary for a series on AI in pathology, covering the basics of generative and traditional machine learning.
2025 · Modern Pathology
Reviews where bias gets into medical AI (the data, the model's development, and how people use it) and the ethical questions from build to deployment.
2024 · Modern Pathology
Reviews how medical AI is regulated: data privacy, software as a medical device, approval and clearance routes, payment, and lab-developed tests.
2025 · Modern Pathology
Compares how generative AI is evaluated (perplexity, BLEU) with how predictive ML is evaluated (AUC, F1, RMSE), for clinicians reading the literature.
2025 · Modern Pathology
Reviews supervised and unsupervised machine learning in medicine and whole-slide image analysis, including explainability and data drift.
2025 · Modern Pathology
The closing review of the series: managing models in clinical use, multimodal and multi-agent AI, and uses in research and teaching.
2024 · Laboratory Investigation
A primer on generating synthetic data (rule-based, ML-based, or hybrid), its uses for training AI and for teaching, and its limits.
2024 · Thrombosis Research
Reviews current and possible uses of AI in venous thromboembolism care, including LLMs and synthetic data, with an emphasis on validation and governance.
2023 · Journal of the American Society of Cytopathology
An editorial on what ChatGPT means for cytopathology, with proposed updates to the journal's rules on responsible AI use.
2020 · Cancer Cytopathology
Compares ways to score a tumor growth marker; augmented reality microscopy with image analysis sped up scoring but wasn't always accurate.
2020 · Cancer Cytopathology
On volunteering with CerviCusco, a Peruvian nonprofit that brings cervical cancer screening to underserved rural communities.
2019 · Respiratory Medicine Case Reports
A case report and review: a seemingly healthy man died suddenly when fluid from a rare cancer of the heart's lining compressed his heart.
Interpretable models on large clinical datasets like NIH All of Us, aimed at questions that come up at the bedside.
Ethan WuInterpretable models trained on patients' earlier diagnoses predicted who develops age-related macular degeneration early. Hypertension, hyperlipidemia, and rheumatoid arthritis stood out, and held up in the NIH All of Us cohort.
Read the paper2026 · Ophthalmology Science
Clustering and interpretable models on All of Us records tie idiopathic epiretinal membrane to systemic conditions like hypertension, hyperlipidemia, and knee osteoarthritis.
2025 · American Journal of Ophthalmology
Models trained on patients' earlier diagnoses predict early-onset AMD. Hypertension, hyperlipidemia, and rheumatoid arthritis stood out and held up in All of Us.
Pairing clinicians with AI models to watch for problems during surgery.
Pairing novice EEG monitors with an AI model held up against expert neurophysiologists at catching brain ischemia during carotid surgery, with fewer false alarms than the AI alone.
Read the paper2026 · medRxiv (preprint)
Pairing novice EEG monitors with an AI model held up against expert neurophysiologists at catching brain ischemia during carotid surgery, with fewer false alarms than the AI alone.
Research questions from the ICU.
This group is just getting going. Want in?
Members bringing their work to the ACOG annual meeting.
Conference abstracts
Abstracts at the ACOG Annual Clinical and Scientific Meeting
Deep learning on retinal OCT scans.
Ethan Wu2025 · PLOS ONE
Benchmarks a diffusion model against leading architectures for segmenting retinal fluid and pigment epithelial detachment on OCT; nnU-Net performed best overall.
Have a clinical question and the data, or a route to it? We have members who want to do the work. Send us a paragraph and we'll find the right people.
Email us with what you're curious about and we'll connect you. The summer course is also a good way in, since it ends with a project of your own.