Diagnostic performance of a deep learning convolutional neural network in the differentiation of combined naevi and melanomas

BACKGROUND

Deep learning convolutional neural networks (CNN) may assist physicians in the diagnosis of melanoma. The capacity of a CNN to differentiate melanomas from combined naevi, the latter representing well-known melanoma simulators, has not been investigated.

OBJECTIVE

To assess the diagnostic performance of a CNN when used to differentiate melanomas from combined naevi in comparison with dermatologists.

METHODS

In this study, a CNN with regulatory approval for the European market (Moleanalyzer-Pro, FotoFinder Systems GmbH, Bad Birnbach, Germany) was used. We attained a dichotomous classification (benign, malignant) in dermoscopic images of 36 combined naevi and 36 melanomas with a mean Breslow thickness of 1.3 mm. Primary outcome measures were the CNN's sensitivity, specificity and the diagnostic odds ratio (DOR) in comparison with 11 dermatologists with different levels of experience.

RESULTS

The CNN revealed a sensitivity, specificity and DOR of 97.1% (95% CI [82.7-99.6]), 78.8% (95% CI [62.8-89.1.3]) and 34 (95% CI [4.8-239]), respectively. Dermatologists showed a lower mean sensitivity, specificity and DOR of 90.6% (95% CI [84.1-94.7]; P = 0.092), 71.0% (95% CI [62.6-78.1]; P = 0.256) and 24 (95% CI [11.6-48.4]; P = 0.1114). Under the assumption that dermatologists use the CNN to verify their (initial) melanoma diagnosis, dermatologists achieve an increased specificity of 90.3% (95% CI [79.8-95.6]) at an almost unchanged sensitivity. The largest benefit was observed in 'beginners', who performed worst without CNN verification (DOR = 12) but best with CNN verification (DOR = 98).

CONCLUSION

The tested CNN more accurately classified combined naevi and melanomas in comparison with trained dermatologists. Their diagnostic performance could be improved if the CNN was used to confirm/overrule an initial melanoma diagnosis. Application of a CNN may therefore be of benefit to clinicians.

Differentiation of combined nevi and melanomas: Case-control study with comparative analysis of dermoscopic features

BACKGROUND AND OBJECTIVES

Combined nevi (CN) show two or more components of major nevus subtypes and simulate melanomas. We investigated a panel of dermoscopic features and three dermoscopic algorithms for differentiating CN from melanomas.

PATIENTS AND METHODS

Retrospective, blinded case-control study using dermoscopic images of 36 CN and 36 melanoma controls. Twenty-one dermoscopic features validated for the diagnosis of melanocytic lesions, the number of colors, and three dermoscopic algorithms were investigated (ABCD rule of dermoscopy, Menzies scoring method, 7-point checklist).

RESULTS

Five of seven features indicative of nevi were observed significantly more frequently in CN than in melanomas (all p < 0.05) and two were exclusively found in CN. Eleven out of 14 features indicative of melanomas were observed significantly more frequently in melanomas than in CN (all p < 0.03) and five were exclusively found in melanomas. The mean (± SD) number of colors in CN was lower than in melanomas (2.1 ± 0.6 versus 3.4 ± 0.7; p < 0.001). Among tested algorithms the ABCD rule of dermoscopy performed best (sensitivity 91.7 %, specificity 77.8 %).

CONCLUSIONS

The ABCD rule of dermoscopy differentiated CN from melanomas most efficiently. Additional knowledge of dermoscopic features to be expected exclusively in either CN or melanomas should help dermatologists to make a correct clinical diagnosis.

Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists

Background

Deep learning convolutional neural networks (CNN) may facilitate melanoma detection, but data comparing a CNN's diagnostic performance to larger groups of dermatologists are lacking.

Methods

Google's Inception v4 CNN architecture was trained and validated using dermoscopic images and corresponding diagnoses. In a comparative cross-sectional reader study a 100-image test-set was used (level-I: dermoscopy only; level-II: dermoscopy plus clinical information and images). Main outcome measures were sensitivity, specificity and area under the curve (AUC) of receiver operating characteristics (ROC) for diagnostic classification (dichotomous) of lesions by the CNN versus an international group of 58 dermatologists during level-I or -II of the reader study. Secondary end points included the dermatologists' diagnostic performance in their management decisions and differences in the diagnostic performance of dermatologists during level-I and -II of the reader study. Additionally, the CNN's performance was compared with the top-five algorithms of the 2016 International Symposium on Biomedical Imaging (ISBI) challenge.

Results

In level-I dermatologists achieved a mean (±standard deviation) sensitivity and specificity for lesion classification of 86.6% (±9.3%) and 71.3% (±11.2%), respectively. More clinical information (level-II) improved the sensitivity to 88.9% (±9.6%, P = 0.19) and specificity to 75.7% (±11.7%, P < 0.05). The CNN ROC curve revealed a higher specificity of 82.5% when compared with dermatologists in level-I (71.3%, P < 0.01) and level-II (75.7%, P < 0.01) at their sensitivities of 86.6% and 88.9%, respectively. The CNN ROC AUC was greater than the mean ROC area of dermatologists (0.86 versus 0.79, P < 0.01). The CNN scored results close to the top three algorithms of the ISBI 2016 challenge.

Conclusions

For the first time we compared a CNN's diagnostic performance with a large international group of 58 dermatologists, including 30 experts. Most dermatologists were outperformed by the CNN. Irrespective of any physicians' experience, they may benefit from assistance by a CNN's image classification.

Clinical trial number

This study was registered at the German Clinical Trial Register (DRKS-Study-ID: DRKS00013570; https://www.drks.de/drks_web/).

Angiogenesis in Ocular and Extraocular Sebaceous Carcinoma

To shed more light on the pathogenesis of sebaceous carcinoma, we analysed the expression of proteins related to angiogenesis in 18 ocular and 22 extraocular sebaceous carcinomas using a broad panel of immunohistochemical markers. To quantify the expression of D2-40, vascular endothelial growth factor, vascular endothelial growth factor receptor-2 and -3, we calculated a quantification score by considering the percentage of positive tumour cells (0=0%, 1=up to 1%, 2=2-10%, 3=11-50%, and 4=>50%) in relation to the staining intensity (0=negative, 1=low, 2=medium, and 3=strong). Additionally, lymphatic microvessel density in the D2-40 stained sections was counted. Vascular endothelial growth factor receptor-3 (quantification score 9.42 ± 2.94) was significantly more strongly expressed than vascular endothelial growth factor receptor-2 (quantification score 2.15 ± 2.42, p < 0.001). Furthermore, epidermal vascular endothelial growth factor expression was negatively correlated with the intratumoural lymphatic vessel density, and the ratio of small lymphatics to large lymphatics was much higher in intratumoural tissue than in paratumoural tissue and in intraindividual control tissue, suggesting a lymphangiogenetic potential of sebaceous carcinoma.

Targeted Therapy in Advanced Melanoma With Rare BRAF Mutations

PURPOSE

BRAF/MEK inhibition is a standard of care for patients with BRAF V600E/K-mutated metastatic melanoma. For patients with less frequent BRAF mutations, however, efficacy data are limited.

METHODS

In the current study, 103 patients with metastatic melanoma with rare, activating non-V600E/K BRAF mutations that were treated with either a BRAF inhibitor (BRAFi), MEK inhibitor (MEKi), or the combination were included. BRAF mutation, patient and disease characteristics, response, and survival data were analyzed.

RESULTS

Fifty-eight patient tumors (56%) harbored a non-E/K V600 mutation, 38 (37%) a non-V600 mutation, and seven had both V600E and a rare BRAF mutation (7%). The most frequent mutations were V600R (43%; 44 of 103), L597P/Q/R/S (15%; 15 of 103), and K601E (11%; 11 of 103). Most patients had stage IV disease and 42% had elevated lactate dehydrogenase at BRAFi/MEKi initiation. Most patients received combined BRAFi/MEKi (58%) or BRAFi monotherapy (37%). Of the 58 patients with V600 mutations, overall response rate to BRAFi monotherapy and combination BRAFi/MEKi was 27% (six of 22) and 56% (20 of 36), respectively, whereas median progression-free survival (PFS) was 3.7 months and 8.0 months, respectively (P = .002). Of the 38 patients with non-V600 mutations, overall response rate was 0% (zero of 15) to BRAFi, 40% (two of five) to MEKi, and 28% (five of 18) to combination treatment, with a median PFS of 1.8 months versus 3.7 months versus 3.3 months, respectively. Multivariable analyses revealed superior survival (PFS and overall survival) with combination over monotherapy in rare V600 and non-V600 mutated melanoma.

CONCLUSION

Patients with rare BRAF mutations can respond to targeted therapy, however, efficacy seems to be lower compared with V600E mutated melanoma. Combination BRAFi/MEKi seems to be the best regimen for both V600 and non-V600 mutations. Yet interpretation should be done with care because of the heterogeneity of patients with small sample sizes for some of the reported mutations.

Physicians' level of hindrance by body hair in dermatoscopy and clinical benefit of an automated hair removal algorithm

BACKGROUND AND OBJECTIVES

Dermatoscopy may be hindered by body hair, and the development of an automated hair removal algorithm (AuHRA) might improve the diagnostic accuracy. However, the physicians' exact level of hindrance and the clinical benefit attained by AuHRA has not been assessed. The objectives of this study are to quantify the physicians' level of hindrance by body hair and the level of improvement in the visibility of underlying dermatoscopic patterns after application of AuHRA to digital images of hair-covered nevi.

PATIENTS AND METHODS

A cross-sectional reader study including 59 sets of dermatoscopic images of benign nevi that were presented to six dermatologists. Each set included three images of one individual nevus (unshaved/physically shaved/digitally shaved with AuHRA), which were compared to each other within each set to assess the level of improvement caused by hair removal.

RESULTS

In comparison to unshaved lesions, dermatologists attributed the highest mean level of improvement to a physical shave (+1.36, p < 0.001) followed by AuHRA's digital shave (+0.79, p < 0.001). The majority of dermatologists considered the application of AuHRA as helpful and confirmed a medical need.

CONCLUSIONS

The dermatologists in our study confirmed a substantial impairment of the dermatoscopic examination by body hair. We demonstrated a clinical benefit attained by AuHRA in comparison to unshaved or physically shaved lesions.

Association Between Surgical Skin Markings in Dermoscopic Images and Diagnostic Performance of a Deep Learning Convolutional Neural Network for Melanoma Recognition

Importance

Deep learning convolutional neural networks (CNNs) have shown a performance at the level of dermatologists in the diagnosis of melanoma. Accordingly, further exploring the potential limitations of CNN technology before broadly applying it is of special interest.

Objective

To investigate the association between gentian violet surgical skin markings in dermoscopic images and the diagnostic performance of a CNN approved for use as a medical device in the European market.

Design and Setting

A cross-sectional analysis was conducted from August 1, 2018, to November 30, 2018, using a CNN architecture trained with more than 120 000 dermoscopic images of skin neoplasms and corresponding diagnoses. The association of gentian violet skin markings in dermoscopic images with the performance of the CNN was investigated in 3 image sets of 130 melanocytic lesions each (107 benign nevi, 23 melanomas).

Exposures

The same lesions were sequentially imaged with and without the application of a gentian violet surgical skin marker and then evaluated by the CNN for their probability of being a melanoma. In addition, the markings were removed by manually cropping the dermoscopic images to focus on the melanocytic lesion.

Main Outcomes and Measures

Sensitivity, specificity, and area under the curve (AUC) of the receiver operating characteristic (ROC) curve for the CNN's diagnostic classification in unmarked, marked, and cropped images.

Results

In all, 130 melanocytic lesions (107 benign nevi and 23 melanomas) were imaged. In unmarked lesions, the CNN achieved a sensitivity of 95.7% (95% CI, 79%-99.2%) and a specificity of 84.1% (95% CI, 76.0%-89.8%). The ROC AUC was 0.969. In marked lesions, an increase in melanoma probability scores was observed that resulted in a sensitivity of 100% (95% CI, 85.7%-100%) and a significantly reduced specificity of 45.8% (95% CI, 36.7%-55.2%, P < .001). The ROC AUC was 0.922. Cropping images led to the highest sensitivity of 100% (95% CI, 85.7%-100%), specificity of 97.2% (95% CI, 92.1%-99.0%), and ROC AUC of 0.993. Heat maps created by vanilla gradient descent backpropagation indicated that the blue markings were associated with the increased false-positive rate.

Conclusions and Relevance

This study's findings suggest that skin markings significantly interfered with the CNN's correct diagnosis of nevi by increasing the melanoma probability scores and consequently the false-positive rate. A predominance of skin markings in melanoma training images may have induced the CNN's association of markings with a melanoma diagnosis. Accordingly, these findings suggest that skin markings should be avoided in dermoscopic images intended for analysis by a CNN.

Trial Registration

German Clinical Trial Register (DRKS) Identifier: DRKS00013570.

Impact of red versus blue light on tolerability and efficacy of PDT: a randomized controlled trial

Various light sources may be used for photodynamic therapy of actinic keratosis since photosensitizing agents are activated by different wavelengths. However, the relative impact of red and blue light irradiation on the efficacy and tolerability of therapy is controversial.

OBJECTIVE

The aim of this study is to compare the efficacy and tolerability of therapy with red versus blue light sources, as well as the patients' evaluation of cosmetic results, clinical response, painfulness and preferred light source for future photodynamic treatments.

METHODS

This is a prospective, single-center, randomized, controlled, open-label study with 28 patients undergoing elective photodynamic therapy.

RESULTS

Red and blue light sources both showed very good results with a complete response rate of 84 % and 85 % respectively. Pain during photodynamic therapy was 6.1 vs. 5.4 (and 2.1 vs. 1.5 eight hours after therapy) on the visual analogue scale. Although these differences were statistically significant, the clinical relevance is low, since the number of therapy interruptions were equally distributed in both groups, and patients' subjective evaluation of the treatment showed no personal preference towards the light sources.

CONCLUSION

Both light sources showed very good clinical results and satisfactory tolerability in this study.

Evaluation of radiotherapeutic and immune-modulatory response to whole brain radiotherapy or stereotactic radiosurgery in patients with brain metastases from malignant melanoma treated with or without ipilimumab (ELEKTRA)

e14104

Background: Immune checkpoint blockers have dramatically improved the survival of patients (pts) suffering from advanced metastasized melanoma. In pts with melanoma brain metastases (MBM) a combination with radiotherapy (RT) is routinely used. Methods: We prospectively included 106 pts with MBM in a non-randomized observational trial with 7 treatment cohorts. Patients in cohort 1-4 were treated with ipilimumab (+/- nivolumab) and either stereotactic (up to 3 MBM) or whole brain RT (≥ 4 MBM) before or after the start of immunotherapy. Cohort 5 and 6 included pts who received RT with an ipilimumab-free systemic treatment and cohort 7 pts were treated with ipilimumab (+/- nivolumab) and no RT. Primary endpoints were immunological response in the peripheral blood (FACS of T cell subsets, ELISpots against melanoma antigens) and radiological response, secondary endpoints were progression free and overall survival. Results: Included pts were in median 61 years old, 72% were male. At trial inclusion, 31% of pts had an elevated LDH. 39% of ipilimumab treated pts received combination therapy with nivolumab. Clinically, ipilimumab treated pts in the early RT groups had better responses of both intra- and extracranial disease (p = 0.04 for both). Multivariate analyses showed a better PFS for pts with early RT (p = 0.02) and normal LDH (p = 0.049). Type of radiation (p = 0.6) and immune therapy (p = 0.8) had no significant influence in this small cohort of pts. Immune monitoring revealed that ipilimumab leads to an increase in activated CD4+ and CD8+ T cells in the peripheral blood which was maintained in responding pts and higher in pts receiving early RT. Treg were not depleted in general but activated by ipilimumab. However, responders displayed a temporary decrease of Treg and activated Treg under treatment. An increase in the detection of melanoma antigens could be observed after 2 cycles of ipilimumab which was higher in pts with combined radioimmunotherapy compared to ipilimumab only. Conclusions: Preliminary data from this small observational trial might lead to a preference of a treatment sequence with radiotherapy first, followed by checkpoint inhibition in pts with MBM.

[Malignant melanoma : Current status]

CLINICAL ISSUE

The incidence of malignant melanoma is continuously increasing. The prognosis of metastatic disease is still limited.

STANDARD TREATMENT

Until a few years ago palliative chemotherapy with a limited response rate was the standard treatment for metastatic melanoma.

TREATMENT INNOVATIONS

Immunotherapy and targeted therapy provide new treatment options. Immune checkpoint inhibitors have significantly improved the prognosis.

DIAGNOSTIC WORK-UP

Regional lymph node sonography, computed tomography (CT) of the neck, chest and abdomen and brain magnetic resonance imaging (MRI) are routinely used. As an alternative to CT scans 18 F fluorodeoxyglucose positron emission tomography (FDG-PET) may be used.

PERFORMANCE AND ACHIEVEMENTS

Immunotherapy provides the chance of long-term disease control in metastatic melanoma. Ipilimumab may provide long-term tumor control in approximately 20% of patients. Median overall survival of approximately 2 years is achieved during therapy with anti-programmed cell death (PD) 1 antibodies. For combined therapy of ipilimumab and nivolumab a response rate of almost 60% is achieved and 2‑year survival is also approximately 60%. The range of immune-mediated side effects demands particular consideration. For response evaluation immune-related response criteria were defined. Furthermore, immunotherapeutic approaches, such as talimogene laherparepvec (T-VEC), which is a modified herpes virus can be used for intralesional injection.

PRACTICAL RECOMMENDATIONS

An individual definition of the appropriate therapy for each patient is of particular importance. In the context of modern therapy regimens close patient monitoring is crucial.