Booth Id:
TMED020T
Category:
Translational Medical Science
Year:
2026
Finalist Names:
Aburadi, Jafar (School: King Abdullah II for Excellence - Muqabilin)
Al-Hayek, Husam (School: Muawiya bin Abi Sufyan Secondary School for Boys)
Abstract:
Every year, 795,000 people die or suffer permanent disability — not from untreatable disease, but from a missed diagnosis. A patient takes a common painkiller alongside a daily prescription, unaware the combination can trigger organ failure or death. A tumor grows silently for months because no tool existed to catch it in time. These are not rare exceptions. They are the daily reality of modern medicine — and they are preventable.
To address this, we developed JomedAI — an integrated agentic AI medical platform. We trained 23 independent CNN-based deep learning models across 13 organ systems using publicly available medical imaging and biomarker datasets, applying EfficientNet-B4, ResNet, and YOLOv11 CNN architectures for 31-class dental panoramic detection. We integrated 1,808 validated clinical scoring models, a drug-drug interaction checker built on DDInter2, and fine-tuned Qwen3-VL-235B via LoRA as MedAI, showing significant improvements on MedQA-USMLE, PubMedQA, and MedMCQA (McNemar's test, p<0.05).
All 23 models were validated on 160,447 independent samples. Three achieved perfect micro-F1=1.000: Tuberculosis X-ray (n=1,000), CT Kidney (n=1,245), Uterine Carcinoma (n=73). Highlights: PCOS=0.999, Hypothyroid=0.9989, Liver Fibrosis=0.9865, Brain Stroke=0.980, Pancreatic Cancer=0.973, Retinal OCT=0.977, Diabetic Retinopathy=0.944 on 90,000 images, Dental YOLO=0.877 across 31 findings. MedAI — an agentic chatbot — discusses results of every tool: diagnostic models, drug interactions, and clinical scores interactively with the physician across multiple turns, rather than simply displaying a number.
Complete. Validated. Deployed across web, desktop, and mobile. JomedAI is not an experiment — it is a medical system built because lives depend on getting this right.
Awards Won: