MedicalNet (Chen et al., Med3D: Transfer Learning for 3D Medical Image Analysis 2019), ported to JAX / Equinox from the Tencent PyTorch release (github.com/Tencent/MedicalNet). A 3D ResNet trunk pretrained on a consortium of 23 medical-imaging segmentation datasets, packaged as a transfer-learning backbone for downstream 3D medical-imaging tasks (segmentation, classification, regression). The published release has seven depths (10 / 18 / 34 / 50 / 101 / 152 / 200) sharing the same architecture template; only the block type (BasicBlock for depths <= 34; Bottleneck for >= 50) and the per-stage block-count vector differ. The stem (7^3 stride-2 conv + BN + ReLU + 3^3 stride-2 maxpool) and four-stage layout (strides 1/2/1/1; dilations 1/1/2/4) are identical across depths. The ilex port ships the headless encoder -- the upstream's conv_seg ConvTranspose+BN+Conv segmentation decoder was a MICCAI-2019 evaluation artefact, not part of the transfer-learning backbone the published weights are designed for. Consumers attach their own task heads.

Bottleneck trunk. Layer4 output is (2048, D/8, H/8, W/8).

Sihong Chen, Kai Ma, Yefeng Zheng (Tencent Healthcare, Youtu Lab)

Chen S., Ma K., Zheng Y. (2019). Med3D: Transfer Learning for 3D Medical Image Analysis. arXiv:1904.00625.

HF Hub license tag: mit

Effective terms: MIT (Tencent Healthcare -- Chen, Ma, Zheng) on both the network code (github.com/Tencent/MedicalNet) and the pretrained weights (mirrored at huggingface.co/TencentMedicalNet/MedicalNet-Resnet*). The ilex JAX / Equinox port code is separately licensed under Apache-2.0 / GPL-3.0.

Upstream license reference: https://opensource.org/licenses/MIT

Network architecture and pretrained weights -- copyright (c) Tencent Healthcare, released under the MIT License. JAX / Equinox port code -- copyright (c) the ilex authors, released under the Apache-2.0 / GPL-3.0 dual license used by ilex itself.

Original weights / reference implementation: https://github.com/Tencent/MedicalNet

This artefact was produced by ilex's save/load pipeline. The architecture is implemented in ilex.models.medical_net.MedicalNet and the weights have been converted from their upstream format. See the upstream source above for the canonical reference.