Consistency of convolutional neural networks in dermoscopic melanoma recognition: A prospective real-world study about the pitfalls of augmented intelligence.

BACKGROUND: Deep-learning convolutional neural networks (CNNs) have outperformed even experienced dermatologists in dermoscopic melanoma detection under controlled conditions. It remains unexplored how real-world dermoscopic image transformations affect CNN robustness. OBJECTIVES: To investigate the consistency of melanoma risk assessment by two commercially available CNNs to help formulate recommendations for current clinical use. METHODS: A comparative cohort study was conducted from January to July 2022 at the Department of Dermatology, University Hospital Basel. Five dermoscopic images of 116 different lesions on the torso of 66 patients were captured consecutively by the same operator without deliberate rotation. Classification was performed by two CNNs (CNN-1/CNN-2). Lesions were divided into four subgroups based on their initial risk scoring and clinical dignity assessment. Reliability was assessed by variation and intraclass correlation coefficients. Excisions were performed for melanoma suspicion or two consecutively elevated CNN risk scores, and benign lesions were confirmed by expert consensus (n = 3). RESULTS: 117 repeated image series of 116 melanocytic lesions (2 melanomas, 16 dysplastic naevi, 29 naevi, 1 solar lentigo, 1 suspicious and 67 benign) were classified. CNN-1 demonstrated superior measurement repeatability for clinically benign lesions with an initial malignant risk score (mean variation coefficient (mvc): CNN-1: 49.5(±34.3)%; CNN-2: 71.4(±22.5)%; p = 0.03), while CNN-2 outperformed for clinically benign lesions with benign scoring (mvc: CNN-1: 49.7(±22.7)%; CNN-2: 23.8(±29.3)%; p = 0.002). Both systems exhibited lowest score consistency for lesions with an initial malignant risk score and benign assessment. In this context, averaging three initial risk scores achieved highest sensitivity of dignity assessment (CNN-1: 94%; CNN-2: 89%). Intraclass correlation coefficients indicated 'moderate'-to-'good' reliability for both systems (CNN-1: 0.80, 95% CI:0.71-0.87, p < 0.001; CNN-2: 0.67, 95% CI:0.55-0.77, p < 0.001). CONCLUSIONS: Potential user-induced image changes can significantly influence CNN classification. For clinical application, we recommend using the average of three initial risk scores. Furthermore, we advocate for CNN robustness optimization by cross-validation with repeated image sets. TRIAL REGISTRATION: ClinicalTrials.gov (NCT04605822).

The role of reflectance confocal microscopy in differentiating melanoma in situ from dysplastic nevi with severe atypia: A cross-sectional study.

BACKGROUND: Melanoma in situ and dysplastic nevi with severe atypia present overlapping histopathologic features. Reflectance confocal microscopy findings can be integrated with the dermatopathology report to improve differentiation between melanoma and dysplastic nevi with severe atypia. OBJECTIVE: To compare prevalence of reflectance confocal microscopy findings between melanoma in situ and dysplastic nevi with severe atypia. METHODS: This retrospective observational study compared reflectance confocal microscopy findings in dermatopathologically diagnosed dysplastic nevi with severe atypia and melanoma in situ, collected between 2007 and 2017 at a private pigmented-lesion clinic. Concordant pathologic diagnosis was defined as unanimous agreement between 3 dermatopathologists who independently reviewed all cases; all other cases were classified as discordant. RESULTS: The study included 112 lesions, 62 concordant melanomas in situ, 28 concordant dysplastic nevi with severe atypia, and 22 discordant lesions. In comparing reflectance confocal microscopy findings in concordant cases, melanoma in situ showed more frequently than dysplastic nevi with severe atypia the presence of epidermal atypical melanocytes as round cells (19/62 vs 0/28; P < .001) and dendritic cells (50/62 vs 6/28; P < .001), as well as a diffuse distribution of epidermal atypical melanocytes (50/54 vs 3/6; P = .002). In contrast, dysplastic nevi with severe atypia showed the presence of dense melanocytic nests more frequently than melanoma in situ did (15/28 vs 14/62; P = .003). LIMITATIONS: The study was based on a limited number of lesions originating from a single clinic. CONCLUSIONS: Reflectance confocal microscopy findings may help differentiate a subset of dysplastic nevi with severe atypia from melanoma in situ.

A retrospective cohort study of the diagnostic value of different subtypes of atypical pigment network on dermoscopy.

BACKGROUND: Atypical network encompasses several patterns. Few studies assess the sensitivity, specificity, and positive and negative predictive values of network subtypes. OBJECTIVE: We assessed the diagnostic value of atypical network subtypes and their histopathologic correlates in cutaneous melanocytic lesions. METHODS: A retrospective search (2014-2018) from a high-risk melanoma clinic for cases scored for atypical network with accompanying dermoscopic photographs yielded 120 lesions (15 melanoma; 30 severely, 38 moderately, and 32 mildly atypical nevi; 4 compound nevi; and 1 junctional nevus). A dermatopathologist blinded to diagnosis assessed dermoscopic and histologic features. Network abnormality correlates with histopathology and clinical diagnoses were assessed with sensitivity, specificity, positive and negative predictive values, and odds ratios. RESULTS: A multivariable model with shiny white streaks (odds ratio 3.02) and inverse network (OR 4.46) was most predictive of melanoma or severe atypia. Positive predictive value for melanoma or severe atypia in decreasing order was inverse network (73.9%), shiny white streaks (71.4%), loss of network (46%), branched streaks (29.4%), and thick brown lines (28.4%). LIMITATIONS: Cases were retrospectively found from a pigmented lesion clinic and evaluated by a single dermatopathologist. CONCLUSION: Shiny white streaks and inverse network are most predictive of melanoma or severe atypia and warrant biopsy if found on dermoscopy.

Improvement of diagnostic confidence and management of equivocal skin lesions by integration of reflectance confocal microscopy in daily practice: Prospective study in 2 referral skin cancer centers.

BACKGROUND: Reflectance confocal microscopy (RCM) allows accurate, noninvasive, in vivo diagnosis for skin cancer. However, its impact on physicians' diagnostic confidence and management is unknown. OBJECTIVES: We sought to assess the physicians' diagnostic confidence and management before and after RCM of equivocal skin lesions. METHODS: Prospective, 2-center, observational study. During clinical practice, 7 dermatologists recorded their diagnostic confidence level (measured in a scale from 0 to 10), diagnosis, and management before and after RCM of clinically/dermoscopically equivocal lesions that raised concern for skin cancer. We also evaluated the diagnostic accuracy before and after RCM. RESULTS: We included 272 consecutive lesions from 226 individuals (mean age, 53.5 years). Diagnostic confidence increased from 6.2 to 8.1 after RCM (P < .001) when RCM confirmed or changed the diagnosis. Lesion management changed in 33.5% cases after RCM (to observation in 51 cases and to biopsy/excision in 31 cases). After RCM, the number needed to excise was 1.2. Sensitivity for malignancy before and after RCM was 78.2% and 85.1%, respectively. Specificity before and after RCM was 78.8% and 80%, respectively. LIMITATIONS: Small sample size, real-life environment, and different levels of expertise among RCM users. CONCLUSION: Physicians' diagnostic confidence and accuracy increased after RCM when evaluating equivocal tumors, frequently resulting in management changes while maintaining high diagnostic accuracy.

Histopathologic correlation of high-risk MelaFindTM lesions: a 3-year experience from a high-risk pigmented lesion clinic.

BACKGROUND: A significant number of pigmented lesions are biopsied to rule out melanoma, but most will be benign. MelaFindTM is a highly sensitive, noninvasive computer-assisted system to aid in clinical diagnosis of melanoma. METHODS: A total of 140 high-risk patients were followed by one expert dermatologist. Biopsies were blindly and independently evaluated by two dermatopathologists and given histologic severity scores (HSS, 0-12) based on the histologic features of melanoma/dysplastic nevi and compared to pathologic diagnoses and MelaFindTM scores. RESULTS: MelaFindTM reduced the biopsy number of clinically ambiguous lesions (923 scanned to 253 biopsied, 73% reduction). Biopsied cases were usually benign (135/253, 53.4%, HSS = 2.8-3.2). Dysplastic nevi with varying degrees of atypia were observed next most commonly (80/253, 31.6%, HSS = 4.7-5.2 for mild dysplasia and 7-7.6 for moderate to severe dysplasia). Melanomas comprised 11/253 (4.3%) of biopsies (HSS = 9.3-10.7). Twenty-four cases were given miscellaneous diagnoses not within the dysplastic nevus-melanoma spectrum (9.5%, HSS = 1.3). Dermal fibrosis was the most commonly identified worrisome histologic feature (177/253, 70%), closely followed by other known atypical features. Nonthreatening histologic features in benign lesions with high MelaFindTM disorganization scores were common. The HSS differed significantly depending on pathologic diagnosis severity, while the MelaFindTM score did not (benign = 2.2; mildly atypical = 4.8; moderately to severely atypical = 2.3; in-situ or invasive melanoma = 3.1). CONCLUSIONS: MelaFindTM unequivocally reduced the number of biopsies, but banal lesions had histologic attributes resulting in high-risk MelaFindTM scores, and MelaFindTM does not correlate with degree of cytologic atypia. Knowledge of these limitations should increase bidirectional confidence when making clinicopathologic correlations in high-risk patients.

Raman biophysical markers in skin cancer diagnosis.

Raman spectroscopy (RS) has demonstrated great potential for in vivo cancer screening; however, the biophysical changes that occur for specific diagnoses remain unclear. We recently developed an inverse biophysical skin cancer model to address this issue. Here, we presented the first demonstration of in vivo melanoma and nonmelanoma skin cancer (NMSC) detection based on this model. We fit the model to our previous clinical dataset and extracted the concentration of eight Raman active components in 100 lesions in 65 patients diagnosed with malignant melanoma (MM), dysplastic nevi (DN), basal cell carcinoma, squamous cell carcinoma, and actinic keratosis. We then used logistic regression and leave-one-lesion-out cross validation to determine the diagnostically relevant model components. Our results showed that the biophysical model captures the diagnostic power of the previously used statistical classification model while also providing the skin's biophysical composition. In addition, collagen and triolein were the most relevant biomarkers to represent the spectral variances between MM and DN, and between NMSC and normal tissue. Our work demonstrates the ability of RS to reveal the biophysical basis for accurate diagnosis of different skin cancers, which may eventually lead to a reduction in the number of unnecessary excisional skin biopsies performed.

Noninvasive diagnosis in dermatology.

In addition to dermoscopy, there are other imaging and biophysical methods for the noninvasive diagnosis of skin lesions. Confocal laser microscopy allows for high-resolution imaging of the epidermis and upper dermis. It is particularly suitable in the differential diagnosis of melanocytic lesions. Optical coherence tomography (OCT) has a lower resolution compared to confocal laser microscopy but a greater depth of penetration. It is primarily used for imaging epithelial skin cancer, especially in the context of monitoring the effectiveness of nonsurgical therapies. Electrical impedance spectroscopy does not yield cutaneous images but rather provides a score based on the cellular irregularity of the skin. Multispectral analysis involves illumination of the skin with different wavelengths and likewise results in the computation of a score. Both methods are used in the differentiation of dysplastic nevi from melanoma. Other diagnostic imaging and biophysical methods are currently still in the developmental stages. By increasing the sensitivity and specificity of clinical and dermoscopic findings, the aforementioned methods bring about an improvement in noninvasive diagnosis. They allow for skin lesions to be monitored over time and therapeutic effects to be quantified. Finally, they facilitate early diagnosis of skin cancer, and help avoid unnecessary surgery of benign lesions.

Compliance with serial dermoscopic monitoring: An academic perspective.

BACKGROUND: For even seasoned practitioners, early melanomas can be difficult to distinguish from melanocytic nevi. Although serial digital dermoscopy is considered by many to be the gold standard for monitoring patients at high risk, poor compliance can seriously alter efficacy. In 2014, a concerning compliance rate of 25% was reported from a single, private clinic. Information is currently limited regarding the determinants of compliance and whether patients at high risk return at an acceptable rate. OBJECTIVE: We sought to determine the compliance rate within the pigmented lesions clinic at our academic institution and identify demographic variables that may influence adherence. METHODS: A retrospective review was conducted using 120 patient charts. RESULTS: An overall compliance rate of 87.5% was observed with 63.3% of patients returning within 1 month of the recommended interval. The most notable risk factor for noncompliance was patient age between 20 and 29 years. Factors promoting adherence include a personal history of melanoma, greater than 5 serially monitored nevi, and a personal history of atypical nevi. LIMITATIONS: The external validity is limited and the sample size is small. CONCLUSION: These findings contradict concerns that adherence to serial monitoring is unacceptably poor and demonstrate that compliance is highest for patients with the greatest inherent risk.

Assessment of tumor thickness in melanocytic skin lesions: comparison of optical coherence tomography, 20-MHz ultrasound and histopathology.

BACKGROUND: Accurate assessment of vertical tumor size is important for surgical treatment planning of melanocytic skin lesions. High-frequency ultrasound (HFUS) is frequently used for this purpose, but overestimation of tumor thickness is known as a problem especially in thin melanocytic lesions. Optical coherence tomography (OCT) as a new imaging technique might be a promising alternative. OBJECTIVE: To evaluate the ability of OCT to accurately determine the vertical tumor thickness of melanocytic skin lesions and to compare it with HFUS and histopathology in order to improve surgical planning. METHODS: In this single-center study, 26 melanocytic lesions were imaged by OCT and HFUS. Vertical lesion dimensions of both methods were compared with histopathological measurements. RESULTS: Bland-Altman plots for OCT and histopathology as well as for HFUS and histopathology revealed better agreement for OCT and histopathology concerning tumor thickness measurements. Tumor thickness values for the melanocytic lesions measured by OCT presented a median tumor thickness of 0.31 mm (range 0.10-0.77) compared to a median tumor thickness of 0.25 mm (range 0.06-1.5) measured by histopathology. The median tumor thickness of HFUS was 0.44 mm (range 0.23-1.1). A Spearman correlation procedure including the correlation coefficient (r) showed a stronger relationship between OCT and histopathology (r = 0.734) compared to HFUS and histopathology (r = 0.390). CONCLUSIONS: On the basis of this smaller study cohort, OCT seems to be more exact than HFUS as far as thickness determination of thin melanocytic skin lesions is concerned.

Atypical familial presentation of FAMMM syndrome with a high incidence of pancreatic cancer: case finding of asymptomatic individuals by EUS surveillance.

BACKGROUND: Pancreatic cancer (PC) is one of the leading causes of cancer death in Western countries. An increased risk for PC is known in a number of hereditary tumor syndromes. In selected individuals at high risk of developing PC, surveillance of the pancreas might be able to detect premalignant lesions and early invasive cancers, and probably improve survival. METHODS: In a Dutch family with atypical phenotypic presentation of the familial atypical multiple mole melanoma syndrome with high incidence of PC related to a mutation in the CDKN2A gene, pancreatic surveillance was offered to asymptomatic gene mutation carriers. RESULTS: Three individuals underwent their first screening with endoscopic ultrasound (EUS) and magnetic resonance imaging at an age of 76, 58, and 51 years. In a mother and a daughter, mass lesions were found by EUS in the tail and body of the pancreas. The smallest lesion was not visualized on subsequent computed tomography and magnetic resonance imaging. After surgical resection histologic examination revealed adenocarcinomas in both cases. The patient with the larger lesion was found to have N1 disease. Side branch intraductal papillary mucinous neoplasias were found in the third patient. CONCLUSIONS: These findings illustrate the potential of the surveillance of high-risk individuals for PC by EUS. Awareness of clinicians of the existence of hereditary syndromes with increased risk for PC may improve identification of high-risk individuals who could benefit from surveillance. Whether screening improves survival remains to be investigated, as is the optimal interval for screening. Side branch intraductal papillary mucinous neoplasias in these patients may serve as a precancerous marker lesion for early intervention to improve survival.

Late occurrence of additional ocular and intracranial pathologies in the linear naevus sebaceous (Feuerstein-Mims) syndrome.

The linear naevus sebaceous syndrome of Feuerstein-Mims belongs to the phacomatoses and classically is composed of a linear cutaneous lesion associated with mental retardation and epilepsy, sometimes also with ocular dystrophies. During an unusually long follow-up of 32 years of such a case, which is reported here, the late occurrence of additional ocular and intracranial pathology are demonstrated: conjunctival and bulbar tumours with progressive deterioration of visual acuity, bilateral intracranial arachnoidal cysts, dilatation and tortuosity of the middle cerebral artery. As a consequence of these observations the prognosis in cases with this syndrome has to be given with special caution.