RadYOLO: Efficient 3D Object Detection and Segmentation for CT and MRI

RadYOLO, an innovative 3D adaptation of YOLO11 specifically designed for medical imaging, has been released in arXiv preprint 2608.00508. This model seeks to mitigate the substantial computational demands associated with deep learning in 3D medical image analysis, providing extensive applicability, impressive detection capabilities, and rapid performance on hardware with limited resources. In tests against nnU-Net and nnDetection across five datasets featuring CT and MRI images with diverse object sizes and frequencies, RadYOLO outperformed nnDetection in four datasets and equaled it in one. While nnU-Net was superior in detecting large organs requiring precise localization, RadYOLO excelled in lesion detection and matched or exceeded nnU-Net's performance in rough localization across all datasets. Additionally, RadYOLO exhibited quicker inference times, although full details remain truncated in the abstract. This advancement holds considerable promise for enhancing efficiency and accessibility in medical imaging diagnostics.