AI4Life Denoising Challenges 2025: Results
by Vera Galinova
The AI4Life Denoising Challenges returned in 2025 with two new tasks: the Microscopy Supervised Denoising Challenge (MDC25) and the Calcium Imaging Denoising Challenge (CIDC25). Both aim to benchmark and improve methods that address noise in microscopy data, a common obstacle for biological and medical imaging.
Why denoising matters
Microscopy is a key tool in life sciences, but image quality is often limited by acquisition noise. This noise can mask fine structures or dynamic processes, making quantitative analysis more difficult. Deep learning–based denoising methods, which learn directly from data rather than relying only on predefined filters, are increasingly used to address this challenge.
The Challenges
Microscopy Supervised Denoising Challenge (MDC25)
This challenge focused on supervised denoising, where models are trained with pairs of noisy and clean images. The setup allowed participants to directly assess how well their methods recover ground truth structures, and to explore strategies for making denoising more precise and consistent across diverse microscopy data. The results can be viewed at https://ai4life-mdc25.grand-challenge.org/results
Calcium Imaging Denoising Challenge (CIDC25)
This task addressed calcium imaging, a widely used technique to record neuronal and cellular activity. Because calcium signals are both spatially and temporally structured, effective denoising needs to preserve not only cell morphology but also the temporal dynamics of activity traces. The challenge provided synthetic datasets with known ground truth to allow controlled evaluation across different noise levels and image content. Participants were also encouraged to develop unsupervised approaches and new evaluation strategies that could be applied to real experimental data, where noise-free ground truth is not available. The challenge is still running!
Participation in 2025
MDC25
- Leaderboard 1: 3 participants / 2 methods
- Leaderboard 2: 6 / 3
- Leaderboard 3: 13 / 4
- Leaderboard 4: 7 / 4
CIDC25
- Preliminary (Content Generalization): 5 / 3
- Final (Content Generalization): 8 / 3
- Preliminary (Noise Level Generalization): 5 / 3
- Final (Noise Level Generalization): 8 / 3
What’s next?
This year’s challenges highlighted how supervised and unsupervised methods can be applied to different types of microscopy data and evaluation settings. While the AI4Life grant has now concluded, the Calcium Imaging Denoising Challenge platform remains accessible, and late submissions are welcome for benchmarking purposes.
👉 https://ai4life-cidc25.grand-challenge.org/