Abstract
Convolutional neural networks can perform well on medical image diagnosis but generally involve heavy computations, which makes them unfeasible in resource-limited scenarios. In this paper, we introduce RAINet, an efficient involution-based model consisting of two designed blocks: Efficient Involution that creates shared group kernels from the pooled input descriptor, and Multi-Scale Involution that combines small and large kernels through a learned gate rather than simple concatenation. Combined with Weight Standardization and Group Normalization for stable small-batch training, these blocks form a ResNet-style architecture optimized for efficiency. Evaluated on four medical imaging datasets against six established baselines, RAINet matches standard high-capacity CNN/ViT accuracy within 0.0052-0.0158, while using as few as 1.28M parameters and 0.5789 GFLOPs. It also shows minor decrease in accuracy when using random weights (0-0.0158), indicating lower dependence on pretraining.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_079.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=ndBmLcZiWh
BibTex
@InProceedings{TasSan_RAINet_MICCAISAT2026,
author = { Tasnim, Sanjida AND Mahtab, MD Ajmain AND Rahman, Rashedur AND Bhuiyan, Siam Tahsin AND Wasi, Sefatul AND Islam, Ashraful AND Alam, Saadia Binte},
title = { { RAINet: Multi-Scale Involutional Model for Resource-Constrained Medical Image Classification Across Modalities } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17264},
month = {pending},
page = {pending}
}
