Quantifying Inflammatory Response and Drug-Aided Resolution in an Atopic Dermatitis Model with Deep Learning.

Noninvasive quantification of dermal diseases aids efficacy studies and paves the way for broader enrollment in clinical studies across varied demographics. Related to atopic dermatitis, accurate quantification of the onset and resolution of inflammatory flare ups in the skin remains challenging because the commonly used macroscale cues do not necessarily represent the underlying inflammation at the cellular level. Although atopic dermatitis affects over 10% of Americans, the genetic underpinnings and cellular-level phenomena causing the physical manifestation of the disease require more clarity. Current gold standards of quantification are often invasive, requiring biopsies followed by laboratory analysis. This represents a gap in our ability to diagnose and study skin inflammatory disease as well as develop improved topical therapeutic treatments. This need can be addressed through noninvasive imaging methods and the use of modern quantitative approaches to streamline the generation of relevant insights. This work reports the noninvasive image-based quantification of inflammation in an atopic dermatitis mouse model on the basis of cellular-level deep learning analysis of coherent anti-Stokes Raman scattering and stimulated Raman scattering imaging. This quantification method allows for timepoint-specific disease scores using morphological and physiological measurements. The outcomes we show set the stage for applying this workflow to future clinical studies.