Crowdsourcing Skin Demarcations of Chronic Graft-Versus-Host Disease in Patient Photographs: Training Versus Performance Study.

BACKGROUND: Chronic graft-versus-host disease (cGVHD) is a significant cause of long-term morbidity and mortality in patients after allogeneic hematopoietic cell transplantation. Skin is the most commonly affected organ, and visual assessment of cGVHD can have low reliability. Crowdsourcing data from nonexpert participants has been used for numerous medical applications, including image labeling and segmentation tasks. OBJECTIVE: This study aimed to assess the ability of crowds of nonexpert raters-individuals without any prior training for identifying or marking cGHVD-to demarcate photos of cGVHD-affected skin. We also studied the effect of training and feedback on crowd performance. METHODS: Using a Canfield Vectra H1 3D camera, 360 photographs of the skin of 36 patients with cGVHD were taken. Ground truth demarcations were provided in 3D by a trained expert and reviewed by a board-certified dermatologist. In total, 3000 2D images (projections from various angles) were created for crowd demarcation through the DiagnosUs mobile app. Raters were split into high and low feedback groups. The performances of 4 different crowds of nonexperts were analyzed, including 17 raters per image for the low and high feedback groups, 32-35 raters per image for the low feedback group, and the top 5 performers for each image from the low feedback group. RESULTS: Across 8 demarcation competitions, 130 raters were recruited to the high feedback group and 161 to the low feedback group. This resulted in a total of 54,887 individual demarcations from the high feedback group and 78,967 from the low feedback group. The nonexpert crowds achieved good overall performance for segmenting cGVHD-affected skin with minimal training, achieving a median surface area error of less than 12% of skin pixels for all crowds in both the high and low feedback groups. The low feedback crowds performed slightly poorer than the high feedback crowd, even when a larger crowd was used. Tracking the 5 most reliable raters from the low feedback group for each image recovered a performance similar to that of the high feedback crowd. Higher variability between raters for a given image was not found to correlate with lower performance of the crowd consensus demarcation and cannot therefore be used as a measure of reliability. No significant learning was observed during the task as more photos and feedback were seen. CONCLUSIONS: Crowds of nonexpert raters can demarcate cGVHD images with good overall performance. Tracking the top 5 most reliable raters provided optimal results, obtaining the best performance with the lowest number of expert demarcations required for adequate training. However, the agreement amongst individual nonexperts does not help predict whether the crowd has provided an accurate result. Future work should explore the performance of crowdsourcing in standard clinical photos and further methods to estimate the reliability of consensus demarcations.

Non-Expert Markings of Active Chronic Graft-Versus-Host Disease Photographs: Optimal Metrics of Training Effects.

Lack of reliable measures of cutaneous chronic graft-versus-host disease (cGVHD) remains a significant challenge. Non-expert assistance in marking photographs of active disease could aid the development of automated segmentation algorithms, but validated metrics to evaluate training effects are lacking. We studied absolute and relative error of marked body surface area (BSA), redness, and the Dice index as potential metrics of non-expert improvement. Three non-experts underwent an extensive training program led by a board-certified dermatologist to mark cGVHD in photographs. At the end of the 4-month training, the dermatologist confirmed that each trainee had learned to accurately mark cGVHD. The trainees' inter- and intra-rater intraclass correlation coefficient estimates were "substantial" to "almost perfect" for both BSA and total redness. For fifteen 3D photos of patients with cGVHD, the trainees' median absolute (relative) BSA error compared to expert marking dropped from 20 cm2 (29%) pre-training to 14 cm2 (24%) post-training. Total redness error decreased from 122 a*·cm2 (26%) to 95 a*·cm2 (21%). By contrast, median Dice index did not reflect improvement (0.76 to 0.75). Both absolute and relative BSA and redness errors similarly and stably reflected improvements from this training program, which the Dice index failed to capture.

Baseline Photos and Confident Annotation Improve Automated Detection of Cutaneous Graft-Versus-Host Disease.

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.