The original model required a high-performance GPU and complex setup. We brought the same deep-learning workflow to ordinary laptops through a lightweight Docker package that runs entirely on CPUs, producing a full diagnostic report from standard MRI data within minutes. This page also hosts a growing Q&A section. Users are invited to share questions, feedback, and experiences to help shape future versions.
Background
The first neural network, developed for research use in an academic setting, ran on a high-performance GPU cluster. While that setup enabled fast training and inference on large MRI datasets, it also created a technical barrier: most hospitals, research groups, and students do not have access to such hardware or the necessary software environment. Even installing the correct versions of PyTorch, MONAI, FSL, and registration tools could be daunting.
To address this, we built u-LINNDA—the user-optimized version that democratizes access to the model. Packaged entirely in Docker, it reproduces the complete workflow on standard CPU-based computers without manual setup or GPU acceleration. The container includes preprocessing (skull stripping, coregistration, and segmentation), inference using the pre-trained DenseNet169 network, and automatic generation of a one-page PDF report summarizing the predicted tumor entity with saliency-map overlays. Typical runtime on a modern laptop (4–8 cores, 16 GB RAM) is under ten minutes per case.
All code and example data are provided under a CC BY-NC 4.0 license for research and teaching purposes. u-LINNDA does not require manual tumor segmentation or specialized imaging sequences, relying instead on standard T1, T1-CE, T2, and FLAIR data. By encapsulating every dependency, the tool ensures full reproducibility and uniform results across systems. While not intended for clinical decision-making, it enables clinicians, engineers, and students to explore AI-assisted neuroradiology safely and transparently.
Community Q&A — Join the conversation
We aim to make this page a living resource for everyone experimenting with u-LINNDA. Over time, we will post frequently asked questions, troubleshooting tips, and user-submitted examples to help others get started or refine their workflows.
If you have questions, ideas, or feedback, please reach out to us — we’ll review submissions regularly and add relevant topics to the public Q&A section. Collaboration and open discussion are central to improving our tool.