Disease Diagnosis From Facial Alternations Using Ensemble CNNs

Identifying health conditions from facial images is crucial for the early detection of certain diseases and provides crucial information for timely intervention. This study introduces a novel ensemble convolutional neural network (CNN) classifier with a visualization technique for diagnosing Bell's palsy, Parry-Romberg syndrome, and Moebius syndrome. As the first step of this study, a dataset was constructed using publicly available images due to the unavailability of benchmark datasets to detect these diseases. The proposed ensemble CNN classifier combines the strengths of ResNet-50, VGG-16, and DenseNet-121 to classify diseases with high accuracy. In addition, a visualization technique was developed to identify the most influential facial regions for detecting these diseases. The proposed ensemble CNN classifier achieved a classification accuracy of 91.96% on the test set of the constructed dataset.