Assessment of facial paralysis (FP) and quantitative grading of facial asymmetry are essential in order to quantify the extent of the condition as well as to follow its improvement or progression. As such, there is a need for an accurate quantitative grading system that is easy to use, inexpensive and has minimal inter-observer variability. A comprehensive automated system to quantify and grade FP is the main objective of this work. An initial prototype has been presented by the authors. The present research aims to enhance the accuracy and robustness of one of this system's modules: the resting symmetry module. This is achieved by including several modifications to the computation method of the symmetry index (SI) for the eyebrows, eyes and mouth. These modifications are the gamma correction technique, the area of the eyes, and the slope of the mouth. The system was tested on normal subjects and showed promising results. The mean SI of the eyebrows decreased slightly from 98.42% to 98.04% using the modified method while the mean SI for the eyes and mouth increased from 96.93% to 99.63% and from 95.6% to 98.11% respectively while using the modified method. The system is easy to use, inexpensive, automated and fast, has no inter-observer variability and is thus well suited for clinical use.
One of the primary concerns of computer‐aided diagnosis is the detection of retinal disorders. The study aims to categorize the patients into choroidal neovascularization, diabetic macular edema, drusen, and normal by using optical coherence tomography (OCT) images. For the first time, two novel transfer learning‐based techniques were used for retinal disorder classification: SqueezeNet and the Inception V3 Net. Two SqueezeNet scenarios were used to compare the performance of the original SqueezeNet and the improved one. A dataset of 11 200 OCT images was used for data partitioning of SqueezeNet and, meanwhile, 18 000 images for Inception V3 Net. The modified SqueezeNet achieved 98% accuracy, a 1.2% improvement over the original. The Inception V3 Net classifier improved its classification accuracy to 98.4%. When compared to other classifiers and a human expert, the transfer learning approach demonstrated its robustness in the challenge of retinal disorders classification with a large dataset.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.