Abstract Dry eye disease (DED) is the most common ocular surface disorder with meibomian gland dysfunction (MGD) as a leading cause. However, precise diagnosis of MGD based on the intrinsic composition remains unattainable, leading to delayed diagnosis and suboptimal therapeutic interventions. This study proposes a high-precision hyperspectral pathological diagnosis scheme for MGD based on a real-time spectral convolutional neural network (SCNN) chip. We analyze the spectral characteristics of meibum, hemoglobin, and its derivatives of meibomian gland pathological sections. The neural network trained on the SCNN chip for MGD diagnosis achieved a diagnostic accuracy of 96.22%, outperforming models based on RGB images of 84.00%. To our knowledge, this is the first study to develop a spectral pathological diagnostic model for MGD and to apply optical neural networks for the diagnosis of MGD. This finding offers new possibilities for effective and accurate MGD management and promotes the performance of SCNN chip for medical applications.
Shi et al. (Mon,) studied this question.