Search

Cutaneous erythema is used in diagnosis and response assessment of cutaneous chronic graft-versus-host disease (cGVHD). The development of objective erythema evaluation methods remains a challenge. We used a pre-trained neural network to segment cGVHD erythema by detecting changes relative to a patient’s registered baseline photo. We fixed this change detection algorithm on human annotations from a single photo pair, by using either a traditional approach or by marking definitely affected (“Do Not Miss”, DNM) and definitely unaffected skin (“Do Not Include”, DNI). The fixed algorithm was applied to each of the remaining 47 test photo pairs from six follow-up sessions of one patient. We used both the Dice index and the opinion of two board-certified dermatologists to evaluate the algorithm performance. The change detection algorithm correctly assigned 80% of the pixels, regardless of whether it was fixed on traditional (median accuracy: 0.77, interquartile range 0.62–0.87) or DNM/DNI segmentations (0.81, 0.65–0.89). When the algorithm was fixed on markings by different annotators, the DNM/DNI achieved more consistent outputs (median Dice indices: 0.94–0.96) than the traditional method (0.73–0.81). Compared to viewing only rash photos, the addition of baseline photos improved the reliability of dermatologists’ scoring. The inter-rater intraclass correlation coefficient increased from 0.19 (95% confidence interval lower bound: 0.06) to 0.51 (lower bound: 0.35). In conclusion, a change detection algorithm accurately assigned erythema in longitudinal photos of cGVHD. The reliability was significantly improved by exclusively using confident human segmentations to fix the algorithm. Baseline photos improved the agreement among two dermatologists in assessing algorithm performance.

To evaluate the effect of illumination, camera orientation, and camera distance on color consistency of different skin tones, several swatches from a Pantone SkinTone Guide deck were selected and photographed at varying distances, illuminations, and angles. The RGB values from each selected swatch from each image were converted to LAB units and compared with the converted “true” values provided by Pantone. The ability of various color references to correct different skin tones was tested by comparing color values from a subject’s forearm skin to “true” color values of the closest Pantone SkinTone swatch found by visual comparison.

Skin biomechanical parameters (dynamic stiffness, frequency, relaxation time, creep, and decrement) measured using a myotonometer (MyotonPRO) could inform the management of sclerotic disease. To determine which biomechanical parameter(s) can accurately differentiate patients with sclerotic chronic graft-versus-host disease from post–hematopoietic cell transplant controls, 15 patients with sclerotic chronic graft-versus-host disease and 11 post–hematopoietic cell transplant controls were measured with the myotonometer on 18 anatomic sites. Logistic regression and two machine learning algorithms (least absolute shrinkage and selection operator regression and random forest) were developed to classify subjects. In univariable analysis, frequency had the highest overfit-corrected area under the curve (0.91). Backward stepwise selection and random forest machine learning identified frequency and relaxation time as the optimal parameters for differentiating patients with sclerotic chronic graft-versus-host disease from post–hematopoietic cell transplant controls. Least absolute shrinkage and selection operator regression selected the combination of frequency and relaxation time (overfit-corrected area under the curve = 0.87). Discriminatory ability was maintained when only the sites accessible while the patient is supine (12 sites) were used. We report the distribution of values for these highly discriminative biomechanical parameters, which could inform the assessment of disease severity in future quantitative biomechanical studies of sclerotic chronic graft-versus-host disease.