Tkaczyk, Eric R.

Objectives/Goals: Manual skin assessment in chronic graft-versus-host disease (cGVHD) can be time consuming and inconsistent (>20% affected area) even for experts. Building on previous work we explore methods to use unmarked photos to train artificial intelligence (AI) models, aiming to improve performance by expanding and diversifying the training data without additional burden on experts. Methods/Study Population: Common to many medical imaging projects, we have a small number of expert-marked patient photos (N = 36, n = 360), and many unmarked photos (N = 337, n = 25,842). Dark skin (Fitzpatrick type 4+) is underrepresented in both sets; 11% of patients in the marked set and 9% in the unmarked set. In addition, a set of 20 expert-marked photos from 20 patients were withheld from training to assess model performance, with 20% dark skin type. Our gold standard markings were manual contours around affected skin by a trained expert. Three AI training methods were tested. Our established baseline uses only the small number of marked photos (supervised method). The semi-supervised method uses a mix of marked and unmarked photos with human feedback. The self-supervised method uses only unmarked photos without any human feedback. Results/Anticipated Results: We evaluated performance by comparing predicted skin areas with expert markings. The error was given by the absolute difference between the percentage areas marked by the AI model and expert, where lower is better. Across all test patients, the median error was 19% (interquartile range 6 – 34) for the supervised method and 10% (5 – 23) for the semi-supervised method, which incorporated unmarked photos from 83 patients. On dark skin types, the median error was 36% (18 – 62) for supervised and 28% (14 – 52) for semi-supervised, compared to a median error on light skin of 18% (5 – 26) for supervised and 7% (4 – 17) for semi-supervised. Self-supervised, using all 337 unmarked patients, is expected to further improve performance and consistency due to increased data diversity. Full results will be presented at the meeting. Discussion/Significance of Impact: By automating skin assessment for cGVHD, AI could improve accuracy and consistency compared to manual methods. If translated to clinical use, this would ease clinical burden and scale to large patient cohorts. Future work will focus on ensuring equitable performance across all skin types, providing fair and accurate assessments for every patient.

AbstractBackgroundThe reliability to non‐invasively identify features of inflammatory dermatoses by reflectance confocal microscopy (RCM) remains unknown. Lack of formal training among RCM readers can result in inconsistent assessments, limiting clinical utility. Specific consensus terminology with representative images is necessary to ensure consistent feature‐level interpretation among RCM readers.Objectives(1) Develop a glossary with representative images of RCM features of cutaneous acute graft‐versus‐host disease (aGVHD) for consistent interpretation among observers, (2) assess the interobserver reproducibility among RCM readers using the glossary, and (3) determine the concordance between RCM and histopathology for aGVHD features.MethodsThrough an iterative process of refinement and discussion among five international RCM experts, we developed a glossary with representative images of RCM features of aGVHD. From April to November 2018, patients suspected of aGVHD were imaged with RCM and subsequently biopsied. 17 lesions from 12 patients had clinically and pathologically confirmed cutaneous aGVHD. For each of these lesions, four dermatopathologists and four RCM readers independently evaluated the presence of aGVHD features in scanned histopathology slides and 1.5 × 1.5 mm RCM submosaics at 4 depths (blockstacks) respectively. RCM cases were adjudicated by a fifth RCM expert. Interobserver reproducibility was calculated by mean pairwise difference (U statistic). Concordance between modalities was determined by fraction of assignments with agreement.ResultsWe present a glossary with representative images of 18 aGVHD features by RCM. The average interobserver reproducibility among RCM readers (75%, confidence interval, CI: 71–79%) did not differ significantly from dermatopathologists (80%, 76–85%). The concordance between RCM and histopathology was 59%.ConclusionsBy using the glossary, the interobserver reproducibility among RCM readers was similar to the interobserver reproducibility among dermatopathologists. There was reasonable concordance between RCM and histopathology to visualize aGVHD features. The implementation of RCM can now be advanced in a variety of inflammatory conditions with a validated glossary and representative image set.

BACKGROUND. Cutaneous erythema and histopathology features in patients post allogeneic hematopoietic cell transplantation (HCT) are nonspecific, making it difficult to distinguish acute graft-versus-host disease (aGVHD) from engraftment syndrome, drug reactions, viral infections, and other etiologies. Biopsies from different locations on the same patient frequently differ in aGVHD grade, and serial biopsies of a single lesion are not possible. Proper biopsy site selection, timing, and serial noninvasive microscopic monitoring may improve our understanding and management of aGVHD. In this study, we tested the feasibility of noninvasive reflectance confocal microscopy (RCM) to detect key histopathology features of cutaneous aGVHD. STUDY POPULATION. We enrolled 11 patients with high clinical suspicion of cutaneous aGVHD, as determined by a transplant physician. In total, 16 lesions of cutaneous aGVHD-affected site were imaged by RCM (6x6 mm2en face images at four different depths) and subsequently biopsied (4x4 mm punch biopsy). 9 out of 11 patients had only skin involvement, and 2 patients had skin and gut aGVHD. 5 patients had >50% body surface area (BSA) affected, 1 had 18-50% BSA, and 5 had <18% BSA. 8 out of 11 patients required systemic corticosteroids and 3 patients received topical corticosteroids. All patients were imaged before starting treatment. METHODS. We used a clinical confocal microscope (Vivascope 1500, Caliber I.D.), which enables real-time evaluation of epidermal and superficial dermal tissue at sub-cellular resolution. Four reflectance confocal microscopists blinded to histopathology independently evaluated the presence or absence of 18 RCM features1. Concurrently, four dermatopathologists blinded to clinical and confocal information determined the presence or absence of 19 histopathology features, as well as the Lerner aGVHD grade. A histologic feature was determined as "present" in a lesion when marked by most experts (expert vote) or resolved by a fifth expert in case of a disagreement among four experts. The reflectance confocal microscopist vote was then correlated to the dermatopathologist expert vote for 17 overlapping features. We also evaluated the interrater variability among microscopists and among dermatopathologists. RESULTS. The main aGVHD features by Lerner grade had > 88% correlation between RCM and histopathology: (1) basal vacuolar change, (2) presence of dyskeratotic keratinocytes, and (3) dermal inflammation. By contrast, dyskeratotic cells and inflammation at the adnexal structures had <50% correlation. This may be attributed to the fact that more adnexal structures are typically visible in the 6x6 mm2 RCM horizontal tissue view compared to vertical section histopathology. We found a similar interrater variability among RCM experts (70%) and dermatopathologists (68%). CONCLUSIONS. In this pilot study, we show that noninvasive label-free RCM of skin enables detection of the main Lerner grade features of cutaneous aGVHD. We found a similar interrater variability among reflectance confocal microscopists and among dermatopathologists who each independently evaluated the presence or absence of a list of aGVHD features. Future studies can build on this work to evaluate the feasibility of RCM to determine cutaneous aGVHD grade, as well as distinguish between rash due to aGVHD and drug reactions. ACKNOWLEDGEMENT. This work was supported in part by Career Development Award Number IK2 CX001785 from the United Sates Department of Veterans Affairs Clinical Science R&D (CSRD) Service. Saknite I, Gill M, Alessi-Fox C, Byrne M, Jagasia M, Gonzalez S, Ardigo M, Tkaczyk ER. Features of cutaneous acute graft-versus-host disease by reflectance confocal microscopy. Br J Dermatol (2019). Disclosures Christi: Caliber I.D.: Employment, Equity Ownership. Byrne:Karyopharm: Research Funding. Jagasia:Kadmon: Consultancy; Incyte: Consultancy; Janssen: Research Funding. Tkaczyk:Incyte: Consultancy.