Weighted Local-Global Facial Feature-Based VMamba Network for Williams Syndrome Diagnosis

Williams Syndrome (WS) is a genetic disorder that is often accompanied by complications such as hypercalcemia, cardiovascular disease, and developmental delays. Early diagnosis of WS is critical for preventing and managing these complications. However, current testing for WS mainly relies on genetic testing, clinical observation, cardiac examination, and blood tests, which can be time-consuming, expensive, and inconvenient. Given that individuals with WS typically exhibit distinct facial features, we propose a new contactless and rapid diagnostic method for WS based on facial image analysis, called Weighted Local-Global Facial Feature-based VMamba Network (WLGFFVMN). Different from the existing facial-based methods for WS diagnosis, WLGFFVMN aims to fully leverage both local and global facial features to obtain more accurate performance. To extract the local salient features of eyes, nose, and mouth, a facial landmark detector is introduced. Then to emphasize the valuable local features while maintaining the global facial features, a pixel-wise weighting unit is designed. Furthermore, we introduce the newly-proposed Visual State Space Model (VMamba) network to serve the WS classification task. Experimental results on our self-constructed WS image dataset demonstrate that our proposed method outperforms other traditional deep convolutional neural networks on the facial-based WS diagnosis tasks.