🚨 DxGPT 🚨

Practical Application of the tool

FDA Plans Phasing Out Animal Testing with AI-Driven Modern Methods

Satya Nadella announced the launch of DxGPT, an AI-powered tool designed to assist healthcare professionals in diagnosing rare diseases and planning treatments.

• The project was created by Julian Isla, a father whose personal experiences with his child’s medical needs inspired him to look for ways to use technology, particularly AI, to make a difference in diagnosing rare diseases, saving precious diagnostic time and alleviating the struggles of both patients and their loved ones.

• Microsoft played a key role in the development of DxGPT, supporting the team with technology and guidance on complying with healthcare regulations.

• The tool leverages large volumes of medical information to help doctors assess symptoms, understand patient histories, and suggest possible diagnoses.

• Clinically evaluated, it achieves 60% accuracy for all diseases, 49% for rare ones.

Let’s break down the use. First case

Case 1: Creutzfeldt-Jakob Disease (Rare Neurological Disease)

65-year-old male with rapidly progressive dementia, myoclonus, visual disturbances. Patients EE’s EEG shows periodic sharp-wave complexes, positive 14-3-3 protein in CSF.

After giving just a brief, and not even EEG findings, DxGPT was able to come up with D/D, where our diagnosis was one of them. It performed really well.

Learn more about DxGPT Platform

Overall, we walk you through the portal, LLM that the model is supported with and global use cases for now.

Case 2: Sporadic Late-Onset Nemaline Myopathy (Rare Neuromuscular Disease)

• Patient: 60-year-old male with Progressive muscle weakness, difficulty rising from chairs, falls, generalized fatigue, and elevated Creatine Kinase. Muscle biopsy showed rod-shaped bodies. (PS: This is a disease I saw and treated. FYI - I have also proudly published on this rare disease.) https://pubmed.ncbi.nlm.nih.gov/31167932/ )

Here, you will notice that I provide DxGPT with an initial history, and it generates 5 different diagnoses. Then I asked to conduct a deeper search, as it was unable to find one of the Diagnoses I was looking for. It still does not reach the diagnosis, or even part of multiple different diagnoses. Then I provide more history of the muscle biopsy, along with the family history, and it conducts further searches, but ultimately still fails to consider “SLONM”.

So we are still a long way from being perfect! A step in the right direction, but we need more of these step-wise advances to reach a point where we can bring transformative change to medical care.

If you enjoyed this, please take a moment to share your review. We hope to bring more of these shortly