With @olegolt.bsky.social & @fraukeb.bsky.social, and a wonderful team of colleagues, we looked into "The utility of explainable A.I. for #MRI analysis"   Preprint: https://doi.org/10.1101/2024.09.27.615357  #compneurosky #NeuroAI #compneuro #neuroscience #neuroskyence #XAI 👇 (1/7) - ThreadSky

About ThreadSky

smnhfmnn.bsky.social • 119 days ago

With @olegolt.bsky.social & @fraukeb.bsky.social, and a wonderful team of colleagues, we looked into

"The utility of explainable A.I. for #MRI analysis"

  Preprint: https://doi.org/10.1101/2024.09.27.615357 

#compneurosky #NeuroAI #compneuro #neuroscience #neuroskyence #XAI

👇 (1/7)

Comments

smnhfmnn.bsky.social•119 days ago

The usage of #AI methods, particularly #deep-learning, became widely popular, as they often provide #SOTA prediction models, such as for #brain-age estimation or disease classification.

💡

However, the model decisions aren’t easy to grasp. (2/7)

smnhfmnn.bsky.social•119 days ago

The models just don't show whether and which biological features are driving their predictions 🧠.

We aimed to close this gap!

We used 3D-convolutional neural networks (#3D-CNN) trained on MR images from 2016 participants (age range 18-82). (3/7)

smnhfmnn.bsky.social•119 days ago

Then, we extracted and related XAI-based relevance maps to features like gray matter volume #GMV, cortical thickness, perivascular spaces (#PVS) and fractional anisotropy #FA, a proxy of the integrity of white matter #WM tracts.

Our results? (4/7)

smnhfmnn.bsky.social•119 days ago

Almost all neuroimaging markers of brain aging, except perivascular spaces (PVS), were highly correlated with XAI-based relevance maps. The strongest correlation was with ventricular volume (r = 0.69)! (5/7)

smnhfmnn.bsky.social•119 days ago

Temporal-parietal cortical thickness, cerebellar gray matter volume, and frontal-occipital white matter tracts showed the strongest correlations with XAI-based relevance. The cerebellum was more important than we expected! 🧠✨

In conclusion, our 3D-CNNs consider a wide range of known (6/7)

smnhfmnn.bsky.social•119 days ago

aging processes for age prediction. Combining deep learning with XAI reveals biologically relevant, multi-feature relationships in the brain. (7/7)

Posting Rules

Be respectful to others
No spam or self-promotion
Stay on topic
Follow Bluesky's terms of service

Comments

Posting Rules

Reply