RSNA Knee Abnormality Detection 2026

The RSNA Knee Abnormality Detection competition has closed: final submissions were due on 22 October 2026 at 11:59 PM UTC (5:29 AM IST on 23 October). The page notes that late submissions are still possible on Kaggle but are not prize-eligible. The Radiological Society of North America (RSNA) runs an AI Challenge on Kaggle, and the page does not announce the next one, so watch Kaggle and RSNA. The online contest is remote, but winners are invited and strongly encouraged to attend the AI Challenge Recognition Event at the RSNA Annual Meeting, which is in person, with the fee waived.

This is a data-science contest, not an application for funding: teams build models that detect twelve clinically important abnormalities on knee MRI scans. The rest of this page is a guide for founders and ML teams preparing for the next RSNA challenge.

What the competition offers

Prizes total US$77,000 (about ₹73 lakh). The main leaderboard pays ten prizes: US$9,000 (about ₹8.6 lakh) for first, then US$7,000, US$6,500, US$6,000 and US$5,500, and US$5,000 each for sixth to tenth. A separate Efficiency Track pays US$7,000, US$6,000 and US$5,000 to the three best submissions on a combined measure of accuracy and speed. A submission can win in both tracks. Winners get the waived-fee invitation to the recognition event. The competition data may be used for commercial and non-commercial purposes under RSNA's licence terms, which is useful for a startup building a product.

Who can enter

Entry is open worldwide to anyone aged 18 or over (or the age of majority where they live, if higher) with a Kaggle account. Residents of Crimea, the so-called Donetsk and Luhansk People's Republics, Cuba, Iran and North Korea, and anyone under US export controls or sanctions, cannot enter. India is not excluded. You may have only one Kaggle account, and teams can have up to five members. Staff of RSNA, Kaggle and their affiliates may take part but cannot win prizes. If you live where prizes cannot legally be awarded, you cannot receive one. It suits ML engineers and teams with experience in medical imaging or multimodal models; it does not suit anyone who cannot work with large image datasets in Kaggle Notebooks.

How to enter

You sign in to Kaggle and accept the rules before the entry deadline, form or merge a team before the team-merger deadline (both were 15 October 2026), and submit predictions as a file with a confidence score for each of twelve labels: ACL, MCL, medial and lateral meniscus, three osteoarthritis labels, effusion, synovitis, Baker's cyst, contusion and fracture. Submissions must go through Kaggle Notebooks with internet access disabled and a run time of up to nine hours on CPU or GPU. You may make five submissions a day. The challenge is the first RSNA AI Challenge dataset that pairs each imaging study with its original radiology report, so models can learn from both the scans and the written findings.

How entries are judged and what happens next

Entries are ranked by the average area under the ROC curve across the twelve labels on a hidden test set. The Efficiency Track combines that score with how long the notebook takes to run, so a lighter model can win even if it is not the most accurate. Winners had until 5 November 2026 to give the host their training code, a short video describing the approach and a method description, and to publish the code and model weights. Winning code is licensed under CC-BY-NC 4.0, which the rules say does not limit commercial use of the code.

About RSNA

RSNA is the Radiological Society of North America. The challenge is organised by a volunteer team of radiologists and researchers from institutions in several countries, and the images and reports were contributed by hospitals and universities in many countries, including Thailand and Türkiye.

Things to check before you enter

Read the rules in full, especially the winners' obligations: you must open-source your solution, and your model is shared for public validation. A practical tip: plan compute early, because everything runs in notebooks with a nine-hour limit and no internet. If you want to attend the recognition event, budget for travel yourself, as the page only waives the event fee. Prize money is cash, but check tax rules for foreign prize income with an adviser. Winners' receipt of prizes also depends on passing the host's review of the solution and eligibility.