245 Deep learning prediction of filaggrin mutation status from palmar images

We previously predicted filaggrin gene (FLG) loss-of-function mutation status from palmar images using textural features (histogram of oriented gradients (HOG) and Haralick). In this study, we used a convolutional neural network (CNN) to predict FLG larger cohort images. Images were available the Tower Hamlets Eczema Assessment study for individuals Bangladeshi origin aged ≤30 years with atopic dermatitis. utilised cropped two regions interest (thenar eminence palm). One image was per patient an 80:10:10 split training, validation, testing. pre-trained CNN (EfficientNetB0). The classification task binary (FLG present vs absent). Model training performed over 30 epochs. keras library in Python 3.9.0 used. included 531 each dataset. For thenar images, area under curve (AUC) 87.8%, accuracy 84.9%, positive predictive value (PPV) 87.5%. AUC 68.4%, 66.0%, PPV 65.9%. higher compared This likely reflects signal-to-noise ratio previous studies having identified hyperlinearity patterns more strongly associated mutations at eminence, including cross-hatch diamond pattern. also achieved feature extraction (84.9% 73.3%). Limitations our dataset size lack external validation. However, minimise risk over-fitting relatively simple as baseline machine learning performance. To facilitate validation have made code available.