Asan Medical Center Researchers Develop Parkinson's Disease GenAI

Researchers from Asan Medical Center (AMC), one of South Korea’s largest hospitals, have developed a generative AI model that autonomously learns and analyses brain scans to diagnose and predict the progression of Parkinson’s disease. The team built a foundational model trained on a large database of dopamine transporter positron emission tomography (DAT PET) images, enabling it to diagnose Parkinson’s, predict its progression, and generate prognostic images. DAT PET imaging is known for its high sensitivity and specificity in detecting dopamine transporters, which helps in early diagnosis of Parkinson’s and related disorders.

The model, named Hierarchical Wavelet Diffusion AutoEncoder (HWDAE), produces a hierarchical representation of these images and was trained on approximately 1,934 DAT PET scans. The researchers later tested the AI to distinguish Parkinson’s from essential tremor, multiple system atrophy, and progressive supranuclear palsy, as well as to predict the onset of motor symptoms.

In a study published in Cell Reports Medicine, the team reported that their AI achieved 99.7% accuracy in distinguishing Parkinson’s from essential tremor and 86.1% accuracy in differentiating it from multiple system atrophy and progressive supranuclear palsy. It also achieved a coefficient of determination of 0.519 in predicting symptom onset, demonstrating notable predictive accuracy. Furthermore, the model maintained its performance when trained and tested with brain scans from other PET devices at AMC and external hospitals.

Beyond improving early-stage diagnosis of Parkinson’s, the AMC AI model also supports clinical decision-making by providing analyses that help medical teams explain disease progression and treatment options to patients. Sun-ju Chung, a study author and professor at AMC’s Department of Neurology, described the work as “a groundbreaking technological advancement that can increase the accuracy of Parkinson’s disease diagnosis” and noted its potential as a practical tool in clinical settings. The team now plans to extend their AI model to other neurodegenerative diseases, according to Professor Kim Nam-guk of AMC’s Department of Convergence Medicine.

This research reflects a broader trend in applying AI to Parkinson’s diagnosis across the Asia-Pacific region. For instance, researchers at RMIT University in Australia developed an AI-driven application that analyses voice recordings to detect the disease, while King Chulalongkorn Memorial Hospital in Thailand tested a similar app that assesses voice, finger movements, tremors, and balance. The Alfred Hospital in Melbourne has also begun studying digital eye movement tests to monitor neurological disorders, including Parkinson’s. Meanwhile, the South Korean government plans to develop a foundational model for assessing cognitive decline as part of its Korean Advanced Research Projects Agency for Health initiatives this year.