▲ (From left) Professor Namkug Kim and Dr. Y.J.Lee of the Department of Convergence Medicine, and Professor Sun Ju Chung of the Department of Neurology at Asan Medical Center

Parkinson’s disease, a progressive neurodegenerative disorder, is often difficult to distinguish from normal aging or other neurological conditions in its early stages, leading to frequent delays in diagnosis. Although DAT PET imaging can assist in diagnosis, it requires specialized personnel and is limited by the subjectivity of interpretation.

A research team led by Professor Namkug Kim and Dr. Y.J.Lee of the Department of Convergence Medicine, and Professor Sun Ju Chung of the Department of Neurology at Asan Medical Center, has developed an artificial intelligence (AI) model that autonomously learns from brain images, generates diagnostic predictions, and accurately identifies Parkinson’s disease. The team trained their AI model on a dataset of 1,934 DAT PET scans. Its performance was evaluated through two key classification tasks: ‘Distinguishing early-stage Parkinson’s disease from essential tremor’, and ‘Differentiating Parkinson’s disease from multiple system atrophy and progressive supranuclear palsy.’ The model achieved impressive diagnostic accuracies of 99.7% and 86.1%, respectively. It also showed potential in predicting the onset of motor symptoms in Parkinson’s disease, achieving an R² value of 0.519, where a value closer to 1 indicates more accurate prediction. The model maintained high performance not only across various PET scanner types within the hospital, but also when applied to external datasets from other institutions.

This AI technology is built upon a foundation model architecture and incorporates a custom-developed hierarchical diffusion model-based encoder. By segmenting brain images into stages and iteratively adding and removing noise, the model achieves highly refined image generation capabilities. In addition to detecting Parkinson’s disease, it is expected to assist in predicting disease progression, thereby supporting prognosis assessment and treatment planning.

The study was published in Cell Reports Medicine, a leading international journal.