Reliability of naevus counts in identifying individuals at high risk of malignant melanoma

Reproducible Naevus Counts Using 3D Total Body Photography and Convolutional Neural Networks

BACKGROUND

The number of naevi on a person is the strongest risk factor for melanoma; however, naevus counting is highly variable due to lack of consistent methodology and lack of inter-rater agreement. Machine learning has been shown to be a valuable tool for image classification in dermatology.

OBJECTIVES

To test whether automated, reproducible naevus counts are possible through the combination of convolutional neural networks (CNN) and three-dimensional (3D) total body imaging.

METHODS

Total body images from a study of naevi in the general population were used for the training (82 subjects, 57,742 lesions) and testing (10 subjects; 4,868 lesions) datasets for the development of a CNN. Lesions were labelled as naevi, or not ("non-naevi"), by a senior dermatologist as the gold standard. Performance of the CNN was assessed using sensitivity, specificity, and Cohen's kappa, and evaluated at the lesion level and person level.

RESULTS

Lesion-level analysis comparing the automated counts to the gold standard showed a sensitivity and specificity of 79% (76-83%) and 91% (90-92%), respectively, for lesions ≥2 mm, and 84% (75-91%) and 91% (88-94%) for lesions ≥5 mm. Cohen's kappa was 0.56 (0.53-0.59) indicating moderate agreement for naevi ≥2 mm, and substantial agreement (0.72, 0.63-0.80) for naevi ≥5 mm. For the 10 individuals in the test set, person-level agreement was assessed as categories with 70% agreement between the automated and gold standard counts. Agreement was lower in subjects with numerous seborrhoeic keratoses.

CONCLUSION

Automated naevus counts with reasonable agreement to those of an expert clinician are possible through the combination of 3D total body photography and CNNs. Such an algorithm may provide a faster, reproducible method over the traditional in person total body naevus counts.

Are we born and do we die without nevi? A cross-sectional study

BACKGROUND

It is well known that the number of nevi varies with age. However, there are a few data in the literature concerning total nevus count at different ages. The aim of this study was to assess and compare the total nevus count among different age groups.

METHODS

The total nevus count was assessed in consecutive patients belonging to nine age groups. Median nevus count was calculated and compared for each age group. Multivariate logistic regression analysis was used to define the influence of sex and phototype on the association between nevus count and age.

RESULTS

We enrolled 900 patients (461, 51.2% females), with a median age of 46 years and a total of 18,136 nevi. Significant differences were found in the median total nevus count among all age groups (P < 0.001), with a significant increase in the first two decades and a significant decrease after the age of 50 years. A similar trend was also observed in subgroups. The differences observed in the total body nevus count among age groups were independent of sex and phototype.

CONCLUSIONS

We demonstrate that the total nevus count significantly increases during childhood and gradually decreases in the elderly.

Accuracy of self-reported nevus and pigmentation phenotype compared with clinical assessment in a population-based study of young Australian adults

BACKGROUND

Awareness of individual risk may encourage improved prevention and early detection of melanoma.

METHODS

We evaluated the accuracy of self-reported pigmentation and nevus phenotype compared with clinical assessment, and examined agreement between nevus counts from selected anatomical regions. The sample included 456 cases with invasive cutaneous melanoma diagnosed between ages 18 to 39 years and 538 controls from the population-based Australian Melanoma Family Study. Participants completed a questionnaire about their pigmentation and nevus phenotype, and attended a dermatologic skin examination.

RESULTS

There was strong agreement between self-reported and clinical assessment of eye color [κ, = 0.78; 95% confidence interval (CI), 0.74-0.81]; and moderate agreement for hair color (κ = 0.46; 95% CI, 0.42-0.50). Agreement between self-reported skin color and spectrophotometer-derived measurements was poor (κ = 0.12; 95% CI, 0.08-0.16) to moderate (Spearman correlation rs = -0.37; 95% CI, -0.32 to -0.42). Participants tended to underestimate their nevus counts and pigmentation; men were more likely to underreport their skin color. The rs was 0.43 (95% CI, 0.38-0.49) comparing clinical total body nevus counts with self-reported nevus categories. There was good agreement between total body nevus counts and site-specific nevus counts, particularly on both arms.

CONCLUSIONS

Young adults have suboptimal accuracy when assessing important risk characteristics including nevus numbers and pigmentation. Measuring nevus count on the arms is a good predictor of full body nevus count.

IMPACT

These results have implications for the likely success of targeted public health programs that rely on self-assessment of these factors.

Development and reliability of a brief skin cancer risk assessment tool

This study aimed to develop and pilot test a brief skin cancer risk assessment tool (BRAT), a self-administered instrument that can be reliably used to assess skin cancer risk. To develop the BRAT, we critically reviewed published literature on risk factors; formulated a draft questionnaire; pilot tested the questionnaire; and retested 1 month later. The BRAT items address the key risk factors for melanoma and other keratinocyte skin cancers: ethnicity, personal and family history of skin cancer, mole count, freckles, childhood residence, sunburn history, and sun sensitivity factors (skin color, natural hair color, ease of sunburning and tanning). One hundred sixty-five persons completed the initial BRAT pilot study, and 52 additional people at moderate- or high-risk completed a second BRAT pilot study. Results were as follows: using a dichotomous risk measure, about 90% of subjects would be correctly classified at baseline and follow-up. Weighted kappa for the total BRAT score (0.41-0.68) and for individual items (0.57-0.99) were fair to good, as were correlation coefficients. The BRAT has acceptable to good reproducibility. Reliability statistics compared favorably with those reported in the literature for similar measures.

The role of screening in the management of skin cancer

The incidence of skin cancer has been rising at an alarming rate for the past several years. This poses a significant public health problem in the United States. Detection and treatment of melanoma early in its course is critical for improved outcome. Of the approaches to cancer control that can reduce mortality from melanoma and nonmelanoma skin cancer, screening holds the greatest promise for a rapid and major impact. Prevention and early detection are crucial in reducing morbidity and mortality from skin cancer. For a number of reasons, however, the full effect of screening for both melanoma and nonmelanoma skin cancers has not been achieved. Controversy exists regarding who should perform screening, who should be screened, and whether screening should be performed at all. It is clear that melanoma and nonmelanoma skin cancer control programs combining primary prevention, education, and screening are in developmental stages. This review will discuss the advantages and disadvantages of screening for skin cancer.

Public health interventions for melanoma. Prevention, early detection, and education

Worldwide melanoma control programs that include some combination of primary prevention, education, and screening activities have only recently begun to undergo an evaluation process. More studies with rigorous design and evaluation are needed. Until then, the proper public health policy guidelines for melanoma control, especially screening, are open to debate. Future studies must determine how screening, early detection, case finding, and education can best be used to reduce mortality and achieve optimal melanoma control.

Discordance in the histopathologic diagnosis of melanoma and melanocytic nevi between expert pathologists

The reliability of a diagnostic test depends on the reproducibility of the result. Many clinical diagnostic tests can be quantified with established ranges and standard deviations. Other tests are more subjective, such as those that depend on analysis of a visual image with an increased possibility of variance in the result. To study this variance, the authors analyzed the performance of expert pathologists in the interpretation of cutaneous melanocytic tumors. A panel of expert pathologists was convened to review anatomic pathology specimens from melanocytic tumors. Each pathologist submitted five specimens, from which 37 were selected for review. Only one slide was used for each case. All specimens were interpreted by each pathologist without consultation with each other. In addition to standard diagnostic terms, each specimen was designated as benign, malignant, or indeterminate. Statistical analysis was used to determine the degree of concordance. The combined kappa statistic for the eight observers and three possible outcomes (benign, malignant, or indeterminate) was 0.50. A kappa statistic of this magnitude, is defined as being moderate. In 62% of the specimens, there was unanimous agreement or only one discordant designation. Thirty-eight percent had two or more discordant interpretations. No single pathologist had a disproportionate number of discordant designations. This study mimics the consultation practice of anatomic pathology and shows the variability and discordance in diagnostic language and designation of biological behavior. The results suggest the criteria for the diagnosis of melanomas and melanocytic nevi need to be refined and more consistently applied.

The influence of history on interobserver agreement for diagnosing actinic keratoses and malignant skin lesions

BACKGROUND

Quantifying interobserver diagnostic agreement is necessary to evaluate skin cancer screening programs, but estimates of variability are incomplete.

OBJECTIVE

We sought to measure agreement between dermatologists for diagnosing actinic keratoses (AKs) and malignant skin lesions and to determine the way in which blinding examiners to patient history affects agreement.

METHODS

We varied the amount of historical information available to examiners in two consecutive patient series (n = 50) presenting to a Veterans Affairs Medical Center dermatology clinic. Two dermatologists examined each patient independently.

RESULTS

Assessing historical features increased the kappa statistic for malignancy recognition from -0.04 to 0.76. kappa Statistics for diagnosing single AKs were 0.17 and 0.15, respectively, and 0.62 and 0.55 for multiple AKs.

CONCLUSION

Agreement was high for diagnosing malignant skin lesions when history was included in the evaluation. Agreement for multiple AKs was higher than for single AKs, although neither was influenced by inclusion of historical features.

The effect of an educational brochure on knowledge and early detection of melanoma

Men over the age of 45 present with thicker, more advanced melanomas than younger people. A randomised trial was conducted in this group to evaluate whether an educational brochure would increase knowledge about melanoma and the ability to recognise and discriminate between pigmented skin lesions. Men in an industrial complex were allocated to an intervention group (n = 110) and two control groups (n = 96 and n = 108). The intervention group was given two educational brochures about melanoma. Their effect on knowledge and ability to detect pigmented lesions was assessed by a questionnaire and a self-examination body chart given before the brochure, and at four weeks and three months after return of the brochure. The control groups did not receive any educational material, but control group 2 received the questionnaire and chart. At the end of the study all participants were examined for pigmented lesions by doctors, whose counts were compared with those of the participants. There was a significant (19.8 per cent) increase in knowledge about melanoma in the intervention group (but not in the control groups), except for discrimination of photos of benign and malignant lesions. The educational material did not improve the ability of those in the intervention group to recognise and count their pigmented lesions nor to discriminate between benign and malignant pigmented lesions. The increased knowledge about melanoma was retained for at least three months.

Self screening for risk of melanoma: validity of self mole counting by patients in a single general practice

OBJECTIVE

To validate self screening by patients of high mole counts, assess the within family association of sun protection behaviour and mole counts, and estimate prevalence of risk factors for melanoma.

SETTING AND SUBJECTS

Systematic sample of families from a single affluent general practice population in Wessex.

DESIGN

Subjects completed a questionnaire about risk factors for melanoma and counted their moles. Subsequently a mole count was done by a general practitioner trained at dermatology clinics.

MAIN OUTCOME MEASURES

Validation of self counts by observer's count. Within family association of sun protection behaviour and mole counts; self reported risk factors.

RESULTS

199/237 subjects (84%) returned the questionnaire; 212/237 (89%) were examined. High counts by patients on the front of the trunk (> 7 moles of > or = 2 mm) were reasonably sensitive (79%), predictive (75%), and specific (97%) of the observer's mole counts (kappa = 0.74), unlike arm or total body counts. Sun protection behaviour correlated between individuals and other family members (Spearman's coefficient r = 0.50, P < 0.01). In the past three months 15/114 adults (13.2%, 95% confidence interval 7.0% to 19.4%) reported any change in a mole and 6/114 (5.3%, 2.0% to 11.1%) "major" changes; 6/109 adults (5.5%, 2.1% to 11.6%) had both high mole counts and freckling.

CONCLUSIONS

Asking patients to count trunk moles could be a feasible way of identifying patients at high risk of melanoma. Concentrating on reported major changes in moles should avoid considerable workload in general practice. The generalisability of these findings and the adverse effects, net benefit in earlier diagnosis and prevention, and workload implications of such self screening need further research.