Multi-site cross-organ calibrated deep learning (MuSClD): Automated diagnosis of non-melanoma skin cancer.

Although deep learning (DL) has demonstrated impressive diagnostic performance for a variety of computational pathology tasks, this performance often markedly deteriorates on whole slide images (WSI) generated at external test sites. This phenomenon is due in part to domain shift, wherein differences in test-site pre-analytical variables (e.g., slide scanner, staining procedure) result in WSI with notably different visual presentations compared to training data. To ameliorate pre-analytic variances, approaches such as CycleGAN can be used to calibrate visual properties of images between sites, with the intent of improving DL classifier generalizability. In this work, we present a new approach termed Multi-Site Cross-Organ Calibration based Deep Learning (MuSClD) that employs WSIs of an off-target organ for calibration created at the same site as the on-target organ, based off the assumption that cross-organ slides are subjected to a common set of pre-analytical sources of variance. We demonstrate that by using an off-target organ from the test site to calibrate training data, the domain shift between training and testing data can be mitigated. Importantly, this strategy uniquely guards against potential data leakage introduced during calibration, wherein information only available in the testing data is imparted on the training data. We evaluate MuSClD in the context of the automated diagnosis of non-melanoma skin cancer (NMSC). Specifically, we evaluated MuSClD for identifying and distinguishing (a) basal cell carcinoma (BCC), (b) in-situ squamous cell carcinomas (SCC-In Situ), and (c) invasive squamous cell carcinomas (SCC-Invasive), using an Australian (training, n = 85) and a Swiss (held-out testing, n = 352) cohort. Our experiments reveal that MuSCID reduces the Wasserstein distances between sites in terms of color, contrast, and brightness metrics, without imparting noticeable artifacts to training data. The NMSC-subtyping performance is statistically improved as a result of MuSCID in terms of one-vs. rest AUC: BCC (0.92 vs 0.87, p = 0.01), SCC-In Situ (0.87 vs 0.73, p = 0.15) and SCC-Invasive (0.92 vs 0.82, p = 1e-5). Compared to baseline NMSC-subtyping with no calibration, the internal validation results of MuSClD (BCC (0.98), SCC-In Situ (0.92), and SCC-Invasive (0.97)) suggest that while domain shift indeed degrades classification performance, our on-target calibration using off-target tissue can safely compensate for pre-analytical variabilities, while improving the robustness of the model.

I need an exact margin measurement for this basal cell carcinoma!

BACKGROUND/OBJECTIVES: Pathology laboratories are required to determine or estimate the measurement uncertainty for all quantitative results, but there is no literature on the uncertainty in margin measurements for skin cancer excisions. METHODS: Six pathologists measured 4-14 histological margins in each of 10 basal cell carcinoma. RESULTS: The mean of measurements from all the margins from all the cases was 1.8 mm (range 0 and 6 mm). Regarding the overall variance in margin measurements across the ten cases, 25% was from variation within cases (differences in margin measurement for a given case, because of different margins and different pathologists measuring each margin, SD 0.7 mm). For a given case, we estimate that 95% of margin measurements would fall approximately within±1.4 mm of the mean measurement for that case. When only pathologists' closest margin for each case were included (for the six cases with uninvolved margins), 6% of the overall variance was from differences within cases (because of different pathologists' measurements of the closest margin, SD 0.2 mm). For a given case without an involved margin, 95% of closest margin measurements would fall approximately within±0.5 mm of the mean closest measurement for that case. CONCLUSIONS: Clinicians should be aware there is uncertainty in reported histological margins.

Assessment of a Diagnostic Classification System for Management of Lesions to Exclude Melanoma.

Importance: The proposed MOLEM (Management of Lesion to Exclude Melanoma) schema is more clinically relevant than Melanocytic Pathology Assessment Tool and Hierarchy for Diagnosis (MATH-Dx) for the management classification of melanocytic and nonmelanocytic lesions excised to exclude melanoma. A more standardized way of establishing diagnostic criteria will be crucial in the training of artificial intelligence (AI) algorithms. Objective: To examine pathologists' variability, reliability, and confidence in reporting melanocytic and nonmelanocytic lesions excised to exclude melanoma using the MOLEM schema in a population of higher-risk patients. Design, Setting, and Participants: This cohort study enrolled higher-risk patients referred to a primary care skin clinic in New South Wales, Australia, between April 2019 and December 2019. Baseline demographic characteristics including age, sex, and related clinical details (eg, history of melanoma) were collected. Patients with lesions suspicious for melanoma assessed by a primary care physician underwent clinical evaluation, dermoscopy imaging, and subsequent excision biopsy of the suspected lesion(s). A total of 217 lesions removed and prepared by conventional histologic method and stained with hematoxylin-eosin were reviewed by up to 9 independent pathologists for diagnosis using the MOLEM reporting schema. Pathologists evaluating for MOLEM schema were masked to the original histopathologic diagnosis. Main Outcomes and Measures: Characteristics of the lesions were described and the concordance of cases per MOLEM class was assessed. Interrater agreement and the agreement between pathologists' ratings and the majority MOLEM diagnosis were calculated by Gwet AC1 with quadratic weighting applied. The diagnostic confidence of pathologists was then assessed. Results: A total of 197 patients were included in the study (102 [51.8%] male; 95 [48.2%] female); mean (SD) age was 64.2 (15.8) years (range, 24-93 years). Overall, 217 index lesions were assessed with a total of 1516 histological diagnoses. Of 1516 diagnoses, 677 (44.7%) were classified as MOLEM class I; 120 (7.9%) as MOLEM class II; 564 (37.2%) as MOLEM class III; 114 (7.5%) as MOLEM class IV; and 55 (3.6%) as MOLEM class V. Concordance rates per MOLEM class were 88.6% (class I), 50.8% (class II), 76.2% (class III), 77.2% (class IV), and 74.2% (class V). The quadratic weighted interrater agreement was 91.3%, with a Gwet AC1 coefficient of 0.76 (95% CI, 0.72-0.81). The quadratic weighted agreement between pathologists' ratings and majority MOLEM was 94.7%, with a Gwet AC1 coefficient of 0.86 (95% CI, 0.84-0.88). The confidence in diagnosis data showed a relatively high level of confidence (between 1.0 and 1.5) when diagnosing classes I (mean [SD], 1.3 [0.3]), IV (1.3 [0.3]) and V (1.1 [0.1]); while classes II (1.8 [0.2]) and III (1.5 [0.4]) were diagnosed with a lower level of pathologist confidence (≥1.5). The quadratic weighted interrater confidence rating agreement was 95.2%, with a Gwet AC1 coefficient of 0.92 (95% CI, 0.90-0.94) for the 1314 confidence ratings collected. The confidence agreement for each MOLEM class was 95.0% (class I), 93.5% (class II), 95.3% (class III), 96.5% (class IV), and 97.5% (class V). Conclusions and Relevance: The proposed MOLEM schema better reflects clinical practice than the MPATH-Dx schema in lesions excised to exclude melanoma by combining diagnoses with similar prognostic outcomes for melanocytic and nonmelanocytic lesions into standardized classification categories. Pathologists' level of confidence appeared to follow the MOLEM schema diagnostic concordance trend, ie, atypical naevi and melanoma in situ diagnoses were the least agreed upon and the most challenging for pathologists to confidently diagnose.

Disordered Eating, Food Insecurity, and Weight Status Among Transgender and Gender Nonbinary Youth and Young Adults: A Cross-Sectional Study Using a Nutrition Screening Protocol.

Purpose: The purpose of this study was to describe the prevalence of and relationships among disordered eating, food insecurity, and weight status among transgender and gender nonbinary youth and young adults. Methods: This cross-sectional study involved a screening protocol to assess disordered eating and food insecurity risk from September to December of 2019 at a gender clinic using five validated measures: (1) previous eating disorder diagnosis (yes/no); (2) Sick, Control, One Stone, Fat, Food Questionnaire (SCOFF); (3) Adolescent Binge Eating Disorder Questionnaire (ADO-BED); (4) Nine-Item Avoidant/Restrictive Food Intake Disorder Screen (NIAS); and (5) Hunger Vital Sign. Age, assigned sex at birth, gender identity, stage of medical transition, and body mass index were collected. Pearson's r correlation coefficients, between-groups t-tests, one-way analysis of variance tests, and Tukey's honest significant difference test were used to characterize the relationships between variables. Results: A total of 164 participants ages 12-23 years completed the screener. Using assigned sex at birth, 1.8% were underweight, 53% were a healthy weight, 17.1% were overweight, and 28.0% were obese. An estimated 8.7% reported a previous eating disorder diagnosis, 28.0% screened positive on the SCOFF, 9.1% on the ADO-BED, 75.0% on the NIAS, and 21.2% on the Hunger Vital Sign. Transgender males scored higher on the NIAS than transgender females (p = 0.03). Those with a previous eating disorder diagnosis scored significantly higher on the Hunger Vital Sign (p < 0.05). Conclusion: Gender clinics should routinely screen for disordered eating, food insecurity, overweight, and obesity to identify patients in need of further evaluation and referral.

Variability in the Histopathological Diagnosis of Non-Melanocytic Lesions Excised to Exclude Melanoma.

INTRODUCTION: The differential diagnosis of lesions excised to exclude melanoma include a variety of benign and malignant melanocytic and non-melanocytic lesions. OBJECTIVES: We examined the variability between pathologists in diagnosing non-melanocytic lesions. METHODS: As part of a larger study prospectively examining the diagnosis of lesions excised to exclude melanoma in 198 patients at a primary care skin cancer clinic in Newcastle, Australia, we compared diagnosis made by 5 experienced dermatopathologists, of 44 non-melanocytic lesions in 44 patients aged 22-90. RESULTS: Forty-four lesions (out of 217 in total) were non-melanocytic. Among the 5 pathologists who examined each case there was marked variability in the terminology used to diagnose each case. The most common variability was found between seborrheic keratosis, large cell acanthoma, solar lentigo, and lichenoid keratosis. The diagnosis made by the majority of the pathologists was deemed to be the reference diagnosis. Versus majority diagnosis, 4% of benign lesions were considered malignant, and 7% of malignant diagnoses were considered as benign. CONCLUSIONS: The different terminology adopted and lack of consensus in the diagnosis of these non-melanocytic lesions in this setting suggests that training AI systems using gold standards may be problematic. We propose a new management classification scheme called MOLEM (Management of Lesions Excised to exclude Melanoma) which expands the previously described MPATH-dx to include non-melanocytic lesions.

Reimagining T Staging Through Artificial Intelligence and Machine Learning Image Processing Approaches in Digital Pathology.

Tumor stage and grade, visually assessed by pathologists from evaluation of pathology images in conjunction with radiographic imaging techniques, have been linked to outcome, progression, and survival for a number of cancers. The gold standard of staging in oncology has been the TNM (tumor-node-metastasis) staging system. Though histopathological grading has shown prognostic significance, it is subjective and limited by interobserver variability even among experienced surgical pathologists. Recently, artificial intelligence (AI) approaches have been applied to pathology images toward diagnostic-, prognostic-, and treatment prediction-related tasks in cancer. AI approaches have the potential to overcome the limitations of conventional TNM staging and tumor grading approaches, providing a direct prognostic prediction of disease outcome independent of tumor stage and grade. Broadly speaking, these AI approaches involve extracting patterns from images that are then compared against previously defined disease signatures. These patterns are typically categorized as either (1) handcrafted, which involve domain-inspired attributes, such as nuclear shape, or (2) deep learning (DL)-based representations, which tend to be more abstract. DL approaches have particularly gained considerable popularity because of the minimal domain knowledge needed for training, mostly only requiring annotated examples corresponding to the categories of interest. In this article, we discuss AI approaches for digital pathology, especially as they relate to disease prognosis, prediction of genomic and molecular alterations in the tumor, and prediction of treatment response in oncology. We also discuss some of the potential challenges with validation, interpretability, and reimbursement that must be addressed before widespread clinical deployment. The article concludes with a brief discussion of potential future opportunities in the field of AI for digital pathology and oncology.

First description of the dermatoscopic features of acquired elastotic hemangioma-a case report.

We present a case of acquired elastotic hemangioma (AEH), a rare, benign vascular tumor. A Caucasian male in his 60s presented with an asymptomatic, solitary, non-pigmented and violaceous lesion of short duration on the dorsum of his hand. The lesion had unique clinical, dermatoscopic and pathological features. Dermatoscopic images of the lesion are presented for characterization and histopathological correlation that have not previously been published or described.