Model and attribution
This browser edition derives from the public MS-PINPOINT MindGlide project and its pinned public checkpoint. The model is not retrained for the browser. It uses FP32 inference, the original 128 × 128 × 64 patches, 1 mm working spacing, 50% overlap and Gaussian blending.
Decoder transposed convolutions are rewritten as pointwise convolutions and spatial shuffles to support browser GPU execution. The rewrite is algebraically equivalent for this network’s non-overlapping upsampling blocks; floating-point kernels can still produce small numerical differences. Export and browser validation evidence is kept with the source code.
Code is released under the repository’s MIT licence. The converted weights are derived from the existing public checkpoint; this conversion does not assign a new licence to the original weights. See the upstream project and model repository for their terms.
Public example
The example is the ICBM 152 Nonlinear Asymmetrical template, version 2009c, 1 mm T1-weighted image, distributed by TemplateFlow. It is a public population template, not a visitor’s scan. Authors: V. Fonov, A. C. Evans, K. Botteron, C. R. Almli, R. C. McKinstry and D. L. Collins. See the template publication.
Copyright (C) 1993–2004 Louis Collins, McConnell Brain Imaging Centre, Montreal Neurological Institute, McGill University.
Permission to use, copy, modify, and distribute this software and its documentation for any purpose and without fee is hereby granted, provided that the above copyright notice appear in all copies. The authors and McGill University make no representations about the suitability of this software for any purpose. It is provided “as is” without express or implied warranty. The authors are not responsible for any data loss, equipment damage, property loss, or injury to subjects or patients resulting from the use or misuse of this software package.
Local processing libraries
ONNX Runtime Web 1.30.0 (MIT), NiiVue 0.69.0 (BSD-2-Clause), Pyodide 0.28.3 (MPL-2.0), NumPy (BSD), SciPy (BSD), scikit-image (BSD), and NiBabel 5.3.2 (MIT). Transitive packages and exact public download hashes are recorded in the source. Runtime distribution licence files are included with the deployment.
Research
MindGlide in Nature Communications. Model outputs need scientific quality control and are not a substitute for clinical assessment.