DaT Parkinson's Challenge

Help support early and accurate detection of parkinsonian syndromes by developing models that classify dopamine transporter (DaT) scans as normal or abnormal. #health

€25,000 in prizes
Completed sep 2026
1,009 joined

Why

Parkinsonian syndromes affect millions of people worldwide. Dopamine transporter (DaT) imaging helps clinicians distinguish neurodegenerative parkinsonian syndromes from other conditions, and reading these scans reliably takes specialized expertise in nuclear medicine.

About one in five DaT scans is difficult to interpret, especially in early-stage or atypical cases. This can delay diagnosis, complicate treatment decisions, and add to the workload of specialist readers. AI tools that read DaT scans accurately and consistently could help expand access to expert-level diagnostic support and accelerate research into neurodegenerative disorders.

The Solution

The goal of the DaT Parkinson's Challenge was to develop computer vision models that classify DaT scans as normal or abnormal. The challenge used a multicenter dataset of DaT scans collected and annotated by French experts across ten hospital centers. Scan resolution and acquisition settings varied from center to center, so models had to generalize across institutions and scanners.

Participants submitted their trained models and inference code, which ran on held-out test scans to predict the probability that each scan was abnormal. Submissions were ranked by log loss.

The Results

More than 1,000 participants joined the challenge, and 373 individuals and teams generated over 2,000 submissions. The winning solutions achieved log loss scores of 0.25-0.27 on the private test set. In terms of area under the ROC curve, all three winning models scored around 0.95.

Bar chart of log loss on the test set for the three winning teams: TheAvengers 0.2532, NavAttack 0.2731, paulonium 0.2732. Lower is better. ROC curves on the test set for the three winning teams, all well above random guessing: TheAvengers AUROC 0.959, NavAttack AUROC 0.952, paulonium AUROC 0.949.


All three winning solutions focused on the striatum, the region where dopamine transporter uptake concentrates and where abnormal scans show reduced and/or asymmetric uptake. Approaches ranged from Siamese neural networks that compare the left and right sides of the striatum to 2D maps of clinical features, and more.

The results suggest that computer vision models can classify DaT scans reasonably accurately across the range of scanners and acquisition settings in the dataset. Tools like these could give specialist readers consistent support, including on scans that are difficult to interpret.

See the results announcement for more information on the winning submissions and the participants who developed them. The winning models will also be indexed in the Bibliothèque Ouverte d'Algorithmes en Santé (BOAS), France's open health algorithm library.


RESULTS ANNOUNCEMENT + MEET THE WINNERS

WINNING MODELS ON GITHUB