MedMLX: On-Device Medical Imaging AI for Apple Silicon

MedMLX, a set of 21 established medical imaging models rewritten for Apple Silicon in MLX, matched original software in 211 of 212 comparisons, including a single brain MRI exception.

• Why it matters: On-device AI could broaden access to multi-model CT/MRI workflows beyond NVIDIA GPUs, but models must run natively on Apple Silicon and reproduce results reliably.
• What they did: The team ported models such as nnU-Net, TotalSegmentator, MedSAM2, and SAM 3, then compared outputs on public cases from 11 datasets against original Intel CPU and NVIDIA GPU software using Dice ≥ 0.99 per segmentation label.
• The result: On an Apple M1 Max (64 GB), with a memory-saving setting, 19 of 21 models used under 16 GB, cutting memory by a median of 49% while increasing processing time by a median of 13%.

AI summary of the abstract · may contain errors · not medical advice · read the paper.