OptiTexture: A Low-Cost, Optical-Deep Learning Solution for Objective Food Texture Assessment in Dysphagia Care

Objective food texture assessment for dysphagia care is crucial for preventing life-threatening aspiration in millions of patients. However, current clinical practice relies on subjective tests that are prone to significant error when performed by non-professionals, creating a pressing need for an objective tool. In this work, we propose a low-cost (sub-$20) multimodal system that objectively assesses food texture by jointly fusing time-domain light scattering (TDLS) and RGB imaging. The proposed system leverages the complementary strengths of the two sensing modalities: TDLS captures subsurface light scattering signatures that reflect food microstructure and mechanical properties such as hardness and particle size, while RGB imaging provides surface geometry and macroscopic visual information. A key challenge arises from the fact that low-cost TDLS hardware produces noisy signals in which texture-related information is easily obscured by chemical and environmental interference. To address this, we introduce a physics-inspired multi-scale convolutional feature extractor tailored to the temporal characteristics of light scattering, combined with a contrastive learning strategy that disentangles micro structural texture cues from confounding factors. Furthermore, to effectively fuse these subsurface scattering signals with surface image features, which may conflict, we propose a physical information attention mechanism. This mechanism can dynamically align the two modalities using texture parameters as a guide, enabling conflict-free fusion. Evaluated on 112 clinically-annotated food samples, our system achieves 91.96% accuracy in classifying The International Dysphagia Diet Standardization Initiative (IDDSI) levels (L3-L6), a 22.32% improvement over a vision-only baseline. It also maintains robust performance under varying ambient conditions like illuminations and temperatures. These results demonstrate a practical solution that enables reliable, objective assessment by non-professionals in home and point-of-care settings.