Unsupervised anomaly appraisal of cleft faces using a StyleGAN2-based model adaptation technique.

A novel machine learning framework that is able to consistently detect, localize, and measure the severity of human congenital cleft lip anomalies is introduced. The ultimate goal is to fill an important clinical void: to provide an objective and clinically feasible method of gauging baseline facial deformity and the change obtained through reconstructive surgical intervention. The proposed method first employs the StyleGAN2 generative adversarial network with model adaptation to produce a normalized transformation of 125 faces, and then uses a pixel-wise subtraction approach to assess the difference between all baseline images and their normalized counterparts (a proxy for severity of deformity). The pipeline of the proposed framework consists of the following steps: image preprocessing, face normalization, color transformation, heat-map generation, morphological erosion, and abnormality scoring. Heatmaps that finely discern anatomic anomalies visually corroborate the generated scores. The proposed framework is validated through computer simulations as well as by comparison of machine-generated versus human ratings of facial images. The anomaly scores yielded by the proposed computer model correlate closely with human ratings, with a calculated Pearson's r score of 0.89. The proposed pixel-wise measurement technique is shown to more closely mirror human ratings of cleft faces than two other existing, state-of-the-art image quality metrics (Learned Perceptual Image Patch Similarity and Structural Similarity Index). The proposed model may represent a new standard for objective, automated, and real-time clinical measurement of faces affected by congenital cleft deformity.

Gauging Facial Abnormality Using Haar-Cascade Object Detector.

The overriding clinical and academic challenge that inspires this work is the lack of a universally accepted, objective, and feasible method of measuring facial deformity; and, by extension, the lack of a reliable means of assessing the benefits and shortcomings of craniofacial surgical interventions. We propose a machine learning-based method to create a scale of facial deformity by producing numerical scores that reflect the level of deformity. An object detector that is constructed using a cascade function of Haar features has been trained with a rich dataset of normal faces in addition to a collection of images that does not contain faces. After that, the confidence score of the face detector was used as a gauge of facial abnormality. The scores were compared with a benchmark that is based on human appraisals obtained using a survey of a range of facial deformities. Interestingly, the overall Pearson's correlation coefficient of the machine scores with respect to the average human score exceeded 0.96.

Toward a Universal Measure of Facial Difference Using Two Novel Machine Learning Models.

A sensitive, objective, and universally accepted method of measuring facial deformity does not currently exist. Two distinct machine learning methods are described here that produce numerical scores reflecting the level of deformity of a wide variety of facial conditions. METHODS: The first proposed technique utilizes an object detector based on a cascade function of Haar features. The model was trained using a dataset of 200,000 normal faces, as well as a collection of images devoid of faces. With the model trained to detect normal faces, the face detector confidence score was shown to function as a reliable gauge of facial abnormality. The second technique developed is based on a deep learning architecture of a convolutional autoencoder trained with the same rich dataset of normal faces. Because the convolutional autoencoder regenerates images disposed toward their training dataset (ie, normal faces), we utilized its reconstruction error as an indicator of facial abnormality. Scores generated by both methods were compared with human ratings obtained using a survey of 80 subjects evaluating 60 images depicting a range of facial deformities [rating from 1 (abnormal) to 7 (normal)]. RESULTS: The machine scores were highly correlated to the average human score, with overall Pearson's correlation coefficient exceeding 0.96 (P < 0.00001). Both methods were computationally efficient, reporting results within 3 seconds. CONCLUSIONS: These models show promise for adaptation into a clinically accessible handheld tool. It is anticipated that ongoing development of this technology will facilitate multicenter collaboration and comparison of outcomes between conditions, techniques, operators, and institutions.