Efficient image classification separates disease from normal anatomy
SCDM: Spatial-Contextual Disentanglement Mamba via Differential Inference for Efficient Image Classification
Computer Vision and Pattern Recognition
Summary
Medical images can be hard to analyze because disease signs are often very similar to normal body parts. The paper’s authors developed a new approach called SCDM that uses two separate pathways in a model: one focuses on picking out disease features, while the other learns what normal anatomy looks like to ignore it. This method improves image classification by making the model better at telling apart sick and healthy areas without needing extra labels or bigger models. They tested this on pneumonia detection, achieving strong results with less computing power.
What this means in practice
- •For medical imaging system developers: Build diagnostic tools that more accurately detect diseases by separating disease features from normal anatomy in medical images using fewer computing resources.
- •For mobile app developers: Create lightweight image analysis apps that identify health conditions on-device by efficiently differentiating abnormal from normal tissue.
Tested on one dataset.
Authors
Mustafa Bora Çelik, Hayriye Aktaş Dinçer, Ayse Keles
Abstract
State Space Models (SSMs), particularly VMamba, have emerged as efficient alternatives for modeling long-range dependencies in medical image analysis. However, distinguishing subtle pathological features from visually similar anatomical backgrounds remains a significant challenge. Existing SSM architectures often learn entangled representations, lacking explicit mechanisms to separate disease-specific signals from normal anatomy. To address this limitation, we propose Spatial-Contextual Differential Mamba (SCDM), an asymmetric dual-branch architecture designed for selective representational disentanglement. SCDM introduces a Positive Branch for extracting discriminative features and a Negative Branch that actively models and suppresses normal anatomical context. This separation is achieved through a similarity-driven repulsion gate and a differential inference rule, which promote competitive feature learning without requiring additional branch labels or increasing model capacity. Evaluated on the RSNA Pneumonia dataset, SCDM achieves competitive classification performance (AUC of 0.858) while requiring significantly fewer parameters (29.4M) and FLOPs (1.44G) compared to standard VMamba and vision transformer baselines. Furthermore, activation analyses demonstrate that our differential mechanism yields highly precise localization, effectively isolating lesions by inhibiting irrelevant anatomical distractors.