Computational diagnostic methods on 2D photographs: A review of the literature

Facial recognition models for identifying genetic syndromes associated with pulmonary stenosis in children

BACKGROUND

Williams-Beuren syndrome, Noonan syndrome, and Alagille syndrome are common types of genetic syndromes (GSs) characterized by distinct facial features, pulmonary stenosis, and delayed growth. In clinical practice, differentiating these three GSs remains a challenge. Facial gestalts serve as a diagnostic tool for recognizing Williams-Beuren syndrome, Noonan syndrome, and Alagille syndrome. Pretrained foundation models (PFMs) can be considered the foundation for small-scale tasks. By pretraining with a foundation model, we propose facial recognition models for identifying these syndromes.

METHODS

A total of 3297 (n = 1666) facial photos were obtained from children diagnosed with Williams-Beuren syndrome (n = 174), Noonan syndrome (n = 235), and Alagille syndrome (n = 51), and from children without GSs (n = 1206). The photos were randomly divided into five subsets, with each syndrome and non-GS equally and randomly distributed in each subset. The proportion of the training set and the test set was 4:1. The ResNet-100 architecture was employed as the backbone model. By pretraining with a foundation model, we constructed two face recognition models: one utilizing the ArcFace loss function, and the other employing the CosFace loss function. Additionally, we developed two models using the same architecture and loss function but without pretraining. The accuracy, precision, recall, and F1 score of each model were evaluated. Finally, we compared the performance of the facial recognition models to that of five pediatricians.

RESULTS

Among the four models, ResNet-100 with a PFM and CosFace loss function achieved the best accuracy (84.8%). Of the same loss function, the performance of the PFMs significantly improved (from 78.5% to 84.5% for the ArcFace loss function, and from 79.8% to 84.8% for the CosFace loss function). With and without the PFM, the performance of the CosFace loss function models was similar to that of the ArcFace loss function models (79.8% vs 78.5% without PFM; 84.8% vs 84.5% with PFM). Among the five pediatricians, the highest accuracy (0.700) was achieved by the senior-most pediatrician with genetics training. The accuracy and F1 scores of the pediatricians were generally lower than those of the models.

CONCLUSIONS

A facial recognition-based model has the potential to improve the identification of three common GSs with pulmonary stenosis. PFMs might be valuable for building screening models for facial recognition. Key messages What is already known on this topic:  Early identification of genetic syndromes (GSs) is crucial for the management and prognosis of children with pulmonary stenosis (PS). Facial phenotyping with convolutional neural networks (CNNs) often requires large-scale training data, limiting its usefulness for GSs. What this study adds:  We successfully built multi-classification models based on face recognition using a CNN to accurately identify three common PS-associated GSs. ResNet-100 with a pretrained foundation model (PFM) and CosFace loss function achieved the best accuracy (84.8%). Pretrained with the foundation model, the performance of the models significantly improved, although the impact of the type of loss function appeared to be minimal. How this study might affect research, practice, or policy:  A facial recognition-based model has the potential to improve the identification of GSs in children with PS. The PFM might be valuable for building identification models for facial detection.

Computer-aided diagnostic screen for Congenital Central Hypoventilation Syndrome with facial phenotype

BACKGROUND

Congenital Central Hypoventilation Syndrome (CCHS) has devastating consequences if not diagnosed promptly. Despite identification of the disease-defining gene PHOX2B and a facial phenotype, CCHS remains underdiagnosed. This study aimed to incorporate automated techniques on facial photos to screen for CCHS in a diverse pediatric cohort to improve early case identification and assess a facial phenotype-PHOX2B genotype relationship.

METHODS

Facial photos of children and young adults with CCHS were control-matched by age, sex, race/ethnicity. After validating landmarks, principal component analysis (PCA) was applied with logistic regression (LR) for feature attribution and machine learning models for subject classification and assessment by PHOX2B pathovariant.

RESULTS

Gradient-based feature attribution confirmed a subtle facial phenotype and models were successful in classifying CCHS: neural network performed best (median sensitivity 90% (IQR 84%, 95%)) on 179 clinical photos (versus LR and XGBoost, both 85% (IQR 75-76%, 90%)). Outcomes were comparable stratified by PHOX2B genotype and with the addition of publicly available CCHS photos (n = 104) using PCA and LR (sensitivity 83-89% (IQR 67-76%, 92-100%).

CONCLUSIONS

Utilizing facial features, findings suggest an automated, accessible classifier may be used to screen for CCHS in children with the phenotype and support providers to seek PHOX2B testing to improve the diagnostics.

IMPACT

Facial landmarking and principal component analysis on a diverse pediatric and young adult cohort with PHOX2B pathovariants delineated a distinct, subtle CCHS facial phenotype. Automated, low-cost machine learning models can detect a CCHS facial phenotype with a high sensitivity in screening to ultimately refer for disease-defining PHOX2B testing, potentially addressing gaps in disease underdiagnosis and allow for critical, timely intervention.

Automatic labeling of facial zones for digital clinical application: An ensemble of semantic segmentation models

INTRODUCTION

The application of artificial intelligence to facial aesthetics has been limited by the inability to discern facial zones of interest, as defined by complex facial musculature and underlying structures. Although semantic segmentation models (SSMs) could potentially overcome this limitation, existing facial SSMs distinguish only three to nine facial zones of interest.

METHODS

We developed a new supervised SSM, trained on 669 high-resolution clinical-grade facial images; a subset of these images was used in an iterative process between facial aesthetics experts and manual annotators that defined and labeled 33 facial zones of interest.

RESULTS

Because some zones overlap, some pixels are included in multiple zones, violating the one-to-one relationship between a given pixel and a specific class (zone) required for SSMs. The full facial zone model was therefore used to create three sub-models, each with completely non-overlapping zones, generating three outputs for each input image that can be treated as standalone models. For each facial zone, the output demonstrating the best Intersection Over Union (IOU) value was selected as the winning prediction.

CONCLUSIONS

The new SSM demonstrates mean IOU values superior to manual annotation and landmark analyses, and it is more robust than landmark methods in handling variances in facial shape and structure.

da Silva C, de Castro Michelassi G, Brentani H, Nunes FL, Machado-Lima A. Computer aided diagnosis of neurodevelopmental disorders and genetic syndromes based on facial images - A systematic literature review

Neurodevelopment disorders can result in facial dysmorphisms. Therefore, the analysis of facial images using image processing and machine learning techniques can help construct systems for diagnosing genetic syndromes and neurodevelopmental disorders. The systems offer faster and cost-effective alternatives for genotyping tests, particularly when dealing with large-scale applications. However, there are still challenges to overcome to ensure the accuracy and reliability of computer-aided diagnosis systems. This article presents a systematic review of such initiatives, including 55 articles. The main aspects used to develop these diagnostic systems were discussed, namely datasets - availability, type of image, size, ethnicities and syndromes - types of facial features, techniques used for normalization, dimensionality reduction and classification, deep learning, as well as a discussion related to the main gaps, challenges and opportunities.

An automatic facial landmarking for children with rare diseases

Two to three thousand syndromes modify facial features: their screening requires the eye of an expert in dysmorphology. A widely used tool in shape characterization is geometric morphometrics based on landmarks, which are precise and reproducible anatomical points. Landmark positioning is user dependent and time consuming. Many automatic landmarking tools are currently available but do not work for children, because they have mainly been trained using photographic databases of healthy adults. Here, we developed a method for building an automatic landmarking pipeline for frontal and lateral facial photographs as well as photographs of external ears. We evaluated the algorithm on patients diagnosed with Treacher Collins (TC) syndrome as it is the most frequent mandibulofacial dysostosis in humans and is clinically recognizable although highly variable in severity. We extracted photographs from the photographic database of the maxillofacial surgery and plastic surgery department of Hôpital Necker-Enfants Malades in Paris, France with the diagnosis of TC syndrome. The control group was built from children admitted for craniofacial trauma or skin lesions. After testing two methods of object detection by bounding boxes, a Haar Cascade-based tool and a Faster Region-based Convolutional Neural Network (Faster R-CNN)-based tool, we evaluated three different automatic annotation algorithms: the patch-based active appearance model (AAM), the holistic AAM, and the constrained local model (CLM). The final error corresponding to the distance between the points placed by automatic annotation and those placed by manual annotation was reported. We included, respectively, 1664, 2044, and 1375 manually annotated frontal, profile, and ear photographs. Object recognition was optimized with the Faster R-CNN-based detector. The best annotation model was the patch-based AAM (p < 0.001 for frontal faces, p = 0.082 for profile faces and p < 0.001 for ears). This automatic annotation model resulted in the same classification performance as manually annotated data. Pretraining on public photographs did not improve the performance of the model. We defined a pipeline to create automatic annotation models adapted to faces with congenital anomalies, an essential prerequisite for research in dysmorphology.

Static and Motion Facial Analysis for Craniofacial Assessment and Diagnosing Diseases

Deviation from a normal facial shape and symmetry can arise from numerous sources, including physical injury and congenital birth defects. Such abnormalities can have important aesthetic and functional consequences. Furthermore, in clinical genetics distinctive facial appearances are often associated with clinical or genetic diagnoses; the recognition of a characteristic facial appearance can substantially narrow the search space of potential diagnoses for the clinician. Unusual patterns of facial movement and expression can indicate disturbances to normal mechanical functioning or emotional affect. Computational analyses of static and moving 2D and 3D images can serve clinicians and researchers by detecting and describing facial structural, mechanical, and affective abnormalities objectively. In this review we survey traditional and emerging methods of facial analysis, including statistical shape modeling, syndrome classification, modeling clinical face phenotype spaces, and analysis of facial motion and affect. Expected final online publication date for the Annual Review of Biomedical Data Science, Volume 5 is August 2022. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.