Automated measurement of hair density and projected apparent width via deep learning-based phototrichogram analysis

Phototrichogram (PTG) imaging enables quantitative assessment of hair density and projected apparent hair shaft width, but manual PTG analysis is labour-intensive and can be sensitive to image quality and operator variability.

We present an automated PTG analysis pipeline that (i) performs hair instance segmentation to count individual hair shafts for density estimation and (ii) quantifies the projected apparent width of each segmented hair shaft using an image-processing module coupled with principal component analysis (PCA) of segmented hair pixel distributions.

Using 259 high-resolution PTG images (6000 × 4000) from 92 patients with predominantly dark pigmented hair acquired in a dermatology clinic, the proposed segmentation model achieved an average precision of 0.914 and recall of 0.870 for hair detection. Method-comparison analyses on the test set demonstrated a moderate-to-strong correlation between automated and clinician-assessed density measurements (r = 0.732, p < 0.001), with a mean bias of -12.08 hairs/cm² and 95% limits of agreement of [-53.30, 29.14]. For projected apparent hair shaft width, the proposed module yielded a mean error of 2.44 μm (4.74%) in average projected apparent hair width compared with clinician measurements.

These results suggest that deep learning-assisted PTG analysis can provide reproducible, time-efficient quantification of key hair parameters for clinical and cosmetic research workflows, particularly for longitudinal monitoring under standardized imaging conditions.