Neural networks for classification and image generation of aging in genetic syndromes

Abstract

In medical genetics, one application of neural networks is the diagnosis of genetic diseases based on images of patient faces. While these applications have been validated in the literature with primarily pediatric subjects, it is not known whether these applications can accurately diagnose patients across a lifespan. We aimed to extend previous works to determine whether age plays a factor in facial diagnosis, as well as to explore other factors that may contribute to the overall diagnosis accuracy. To investigate this, we chose two relatively common conditions, Williams syndrome and 22q11.2 deletion syndrome. We built a neural network classifier trained on images of affected and unaffected individuals of different ages. Our classifier outperformed clinical geneticists at recognizing face images of these two conditions within each of the age groups (the performance varied between the age groups): (1) under 2 years old, (2) 2-9 years old, (3) 10-19 years old, (4) 20-34 years old, and (5) ≥35 years old. The overall accuracy improvement by our classifier over the clinical geneticists was 15.5% and 22.7% for Williams syndrome and 22q11.2 deletion syndrome, respectively. Additionally, comparison of saliency maps revealed that key facial features learned by the neural network differed slightly with respect to age. Finally, joint training real images with multiple different types of fake images created by a generative adversarial network showed up to 3.25% accuracy gain in classification accuracy.

Approximating facial expression effects on diagnostic accuracy via generative AI in medical genetics

“…Despite these limitations, one overall interesting conclusion, which echoes some of our previous work, is that a DL model and humans simply perform differently under different circumstances ( Duong et al , 2023 ). As AI tools are adopted in clinical scenarios, better understanding of these differences will be important to ensure that they are useful in helping to diagnose and care for people affected by genetic conditions.…”

“…After calibration, each participant viewed the 32 images for 7 s per image. We chose 7 s for the viewing time after extensive initial testing and per our previous work using eye-tracking in a related way, as subjective feedback and preliminary assessments showed that this amount of time was sufficient to assess an image but minimized participants visually revisiting areas of the image in a way that might not inform the assessment ( Duong et al , 2023 ).…”