An automated framework to classify skin lesions using Multi-Head Self Attention Layer-based Vision Transformers

The medical sciences are facing a major problem with the auto-detection of disease due to the fast growth in population density. Intelligent systems assist medical professionals in early disease detection and also help to provide consistent treatment that reduces the mortality rate. Skin cancer is considered to be the deadliest and most severe kind of cancer. Medical professionals utilize dermoscopy images to make a manual diagnosis of skin cancer. This method is labor-intensive and time-consuming and demands a considerable level of expertise. Automated detection methods are necessary for the early detection of skin cancer. The occurrence of hair and air bubbles in dermoscopic images affects the diagnosis of skin cancer. This research aims to classify eight different types of skin cancer, namely actinic keratosis (AKs), dermatofibroma (DFa), melanoma (MELa), basal cell carcinoma (BCCa), squamous cell carcinoma (SCCa), melanocytic nevus (MNi), vascular lesion (VASn), and benign keratosis (BKs). In this study, we propose SNC_Net, which integrates features derived from dermoscopic images through deep learning (DL) models and handcrafted (HC) feature extraction methods with the aim of improving the performance of the classifier. A convolutional neural network (CNN) is employed for classification. Dermoscopy images from the publicly accessible ISIC 2019 dataset for skin cancer detection is utilized to train and validate the model. The performance of the proposed model is compared with four baseline models, namely EfficientNetB0 (B1), MobileNetV2 (B2), DenseNet-121 (B3), and ResNet-101 (B4), and six state-of-the-art (SOTA) classifiers. With an accuracy of 97.81%, a precision of 98.31%, a recall of 97.89%, and an F1 score of 98.10%, the proposed model outperformed the SOTA classifiers as well as the four baseline models. Moreover, an Ablation study is also performed on the proposed method to validate its performance. The proposed method therefore assists dermatologists and other medical professionals in early skin cancer detection.

Skin cancer is one of the most common cancers, and its early detection can have a huge impact on its outcomes. Deep learning, especially convolutional neural networks, perform well in processing massive amounts of data, especially image data in classifying skin cancer. In this paper, convolutional neural networks are mainly used to diagnose and classify 7 types of skin lesions, including melanoma, basal cell carcinoma, melanocytic nevus, actinic keratosis, and intraepithelial carcinoma, benign keratinoid lesions, dermatofibroma, and vascular lesions. First, the characteristics of skin lesion images are analyzed, using image processing technology and sampling technology to preprocess skin lesion images. Then the training parameters of imageNet network are adjusted through the idea of transfer learning on InceptionV3, ResNet50, DenseNet201, and other networks to perform training classification. Furthermore, different convolutional neural network models are built for classification. In order to validate the classification performance of various convolutional neural network models, this paper adopts ISIC 2017 HAM10000 dataset for experiments. The experimental results show that proper preprocessing is necessary when applying CNN for image classification. In classifying the 224*224 skin lesion images, the classical deep convolutional network with DenseNet201 achieved a remarkable performance classification accuracy of 99.12% for training and 86.91% for testing.

Aquaporin 3 (AQP3) is the membrane channel of water and involved in fluid homeostasis. The aim of this study was to reveal the expression and significance of AQP3 in cutaneous lesions. We analyzed AQP3 mRNA levels using RT-PCR in 311 cutaneous lesions and confirmed AQP3 expression in these lesions by immunohistochemistry. AQP3 mRNA was detected in normal epidermis, seborrheic keratosis, solar keratosis, Bowen's disease, squamous cell carcinoma, eccrine poroma, apocrine carcinoma, and sebaceoma; however, AQP3 mRNA was absent in basal cell carcinoma, nevocellular nevus, or malignant melanoma. By immunohistochemistry, diffuse AQP3 expression was seen in all keratotic lesions including seborrheic keratosis, verruca vulgaris, molluscum contagiosum, solar keratosis, Bowen's disease, and squamous cell carcinoma. Diffuse AQP3 expression was also present in all extramammary Paget's disease. No AQP3 staining was obtained in basal cell carcinoma. Positive AQP3 staining was seen in sweat gland tumors including hidradenoma, eccrine poroma, and apocrine carcinoma. Among sebaceous tumors, AQP3 expressed diffusely in all sebaceous hyperplasia and sebaceous adenoma, but not in sebaceous carcinomas. Only focal AQP3 staining was seen in nevocellular nevus and no AQP3 staining in melanoma. Our findings indicate the function of AQP3 maintained in most skin tumors. AQP3 may be used for differential diagnosis in skin tumors.

A machine learning approach for skin disease detection and classification using image segmentation

Concern about skin cancer is a common reason for people from predominantly fair-skinned populations to present to primary care doctors. To examine the frequency and body-site distribution of malignant, pre-malignant and benign pigmented skin lesions excised in primary care. This prospective study conducted in Queensland, Australia, included 154 primary care doctors. For all excised or biopsied lesions, doctors recorded the patient's age and sex, body site, level of patient pressure to excise, and the clinical diagnosis. Histological confirmation was obtained through pathology laboratories. Of 9650 skin lesions, 57·7% were excised in males and 75·0% excised in patients ≥ 50 years. The most common diagnoses were basal cell carcinoma (BCC) (35·1%) and squamous cell carcinoma (SCC) (19·7%). Compared with the whole body, the highest densities for SCC, BCC and actinic keratoses were observed on chronically sun-exposed areas of the body including the face in males and females, the scalp and ears in males, and the hands in females. The density of BCC was also high on intermittently or rarely exposed body sites. Females, younger patients and patients with melanocytic naevi were significantly more likely to exert moderate/high levels of pressure on the doctor to excise. More than half the excised lesions were skin cancer, which mostly occurred on the more chronically sun-exposed areas of the body. Information on the type and body-site distribution of skin lesions can aid in the diagnosis and planned management of skin cancer and other skin lesions commonly presented in primary care.

Background Elastic scattering spectroscopy (ESS) is an optical biopsy technique that can distinguish between a normal and abnormal tissue in vivo without the need to remove it. The handheld device measures ESS spectra of skin lesions and classifies lesions as either malignant or benign with an output of “Investigate Further” or “Monitor,” respectively, with positive results accompanied by a spectral score output from 1 to 10, indicating how similar the lesion is to the malignant lesions the device was trained on. The algorithm was trained and validated with over 11,000 spectral scans from over 3500 skin lesions. Objective The purpose of this study was to evaluate the safety and effectiveness of the handheld ESS device in detecting the most common types of skin cancer. Methods A prospective, single-arm, investigator-blinded, multicenter study conducted at 4 investigational sites in the United States was performed. Patients who presented with skin lesions suggestive of melanoma, basal cell carcinoma, squamous cell carcinoma, and other highly atypical lesions were evaluated with the handheld ESS device. A validation performance analysis was performed with 553 lesions from 350 subjects with algorithm version 2.0. An independent test set of 281 lesions was selected and used to evaluate device performance in the detection of melanoma, basal cell carcinoma (BCC), and squamous cell carcinoma (SCC). Statistical analyses included overall effectiveness analyses for sensitivity and specificity as well as subgroup analyses for lesion diagnoses. Results The overall sensitivity of the device was 92.3% (95% CI: 87.1 to 95.5%). The sensitivity for subgroups of lesions was 95% (95% CI 75.1% to 99.9%) for melanomas, 94.4% (95% CI 86.3% to 98.4%) for BCCs, and 92.5% (95% CI 83.4% to 97.5%) for SCCs. The overall device specificity was 36.6% (95% CI 29.3% to 44.6%). There was no statistically significant difference between the dermatologist performance and the ESS device (P=.2520). The specificity of the device was highest for benign melanocytic nevi (62.5%) and seborrheic keratoses (78.2%). The overall positive predictive value (PPV) was 59.8%, and the negative predictive value (NPV) was 81.9% with the study’s malignancy prevalence rate of 51%. For a prevalence rate of 5%, the PPV was estimated to be 7.1%, and the NPV was estimated to be 98.9%. For a prevalence rate of 7%, the PPV was estimated to be 9.8%, and the NPV was estimated to be 98.4%. For a prevalence rate of 15%, the PPV was estimated to be 20.3%, and the NPV was 96.4%. Conclusions The handheld ESS device has a high sensitivity for the detection of melanoma, BCC, and SCC. Coupled with clinical exam findings, this device can aid physicians in detecting a variety of skin malignancies. The device output can aid teledermatology evaluations by helping frontline providers determine which lesions to share for teledermatologist evaluation as well as potentially benefitting teledermatologists’ virtual evaluation, especially in instances of suboptimal photo quality. Acknowledgments This study was sponsored by Dermasensor Inc. Conflicts of Interest None declared.

Melanoma is deadly skin cancer. There is a high similarity between different kinds of skin lesions, which lead to incorrect classification. Accurate classification of a skin lesion in its early stages saves human life. In this paper, a highly accurate method proposed for the skin lesion classification process. The proposed method utilized transfer learning with pre-trained AlexNet. The parameters of the original model used as initial values, where we randomly initialize the weights of the last three replaced layers. The proposed method was tested using the most recent public dataset, ISIC 2018. Based on the obtained results, we could say that the proposed method achieved a great success where it accurately classifies the skin lesions into seven classes. These classes are melanoma, melanocytic nevus, basal cell carcinoma, actinic keratosis, benign keratosis, dermatofibroma, and vascular lesion. The achieved percentages are 98.70%, 95.60%, 99.27%, and 95.06% for accuracy, sensitivity, specificity, and precision, respectively.

Distribuição dos diagnósticos de lesões pré-neoplásicas e neoplásicas de pele no Hospital Universitário Evangélico de Curitiba

Melanoma, one of the most deadly skin lesions is a major cause of death if not detected at earlier stages. Due to the high similarity between melanoma and other similar lesions, the investigation and diagnosis of these lesions are complex and time consuming job. To address such problems, in this paper, we present a machine learning integrated automated lesion classification in Healthcare Cyber-Physical System (HCPS) or smart healthcare framework. We propose an automated classification of various types of skin lesions by using deep learning tools, which save time, human effort, money, and human life by diagnosing melanoma at the earlier stage. Our model can classify 7 different types of skin lesions that are Melanocytic nevi, Melanoma, Benign keratosis, Basal cell carcinoma, Actinic keratoses, Vascular lesions, Dermatofibroma. The purpose of this paper is to classify the skin lesion with our pre-trained model. We have trained our model on multiple platforms like DenseNet169, DenseNet121, ResNet50, ResNet152 and found DenseNet 169 is most accurate among all. Therefore, the proposed technique may be useful for diagnosis of melanoma in rural places, where healthcare facilities, availability of oncologists and doctors are limited.

Dermatological problems are the most widely spread skin diseases amongst human beings. They can be infectious, chronic, and sometimes may also lead to serious health problems such as skin cancer. Generally, rural area clinics lack trained dermatologists and mostly rely on the analysis of remotely accessible experts through mobile-based networks for sharing the images and other related information. Under such circumstances, poor image quality introduced due to the capturing device results in misleading diagnosis. Here, a genetic-algorithm- (GA-) based approach used as an image enhancement technique has been explored to improve the low quality of the dermatological images received from the rural clinic. The diagnosis is performed on the enhanced images using convolutional neural network (CNN) classifier for the identification of the diseases. The scope of this paper is limited to only motion blurred images, which is the most prevalent problem in capturing of the images, specifically when any of the two (device or the object) may move unpredictably. Seven types of skin diseases, namely, melanoma, melanocytic nevus, basal cell carcinoma, actinic keratosis, benign keratosis, vascular lesion, and squamous cell carcinoma, have been investigated using ResNet-152 giving an overall accuracy of 87.40% for the blurred images. Use of GA-enhanced images increased the accuracy to 95.85%. The results were further analyzed using a confusion matrix and t-test-based statistical investigations. The advantage of the proposed technique is that it reduces the analysis time and errors due to manual diagnosis. Furthermore, speedy and reliable diagnosis at the earliest stage reduces the risk of developing more severe skin problems.

200 Risks and timing of future skin cancers in patients with actinic keratosis

The activation of oncogenes is an important step in tumorigenesis, and recently, oncogene-induced senescence (OIS) was proposed as a critical barrier against malignant transformation in normal primary cells. The aim of this study was to examine the activation of fibroblast growth factor receptor 3 (FGFR3) as an oncogene product and OIS in human skin tumours. We investigated the activation of FGFR3 and OIS by mutation and immunohistochemical analysis in skin tumours, including seborrhoeic keratosis, actinic keratosis (AK), Bowen's disease (BD), basal cell carcinoma (BCC) and squamous cell carcinoma (SCC). Activated point mutations of FGFR3 were identified in four of 22 cases (18%) of seborrhoeic keratosis, but no mutation was detected in the other skin tumours. Twenty-seven of 31 cases (87%) of seborrhoeic keratosis showed moderately to strongly positive expression of the FGFR3 protein, but almost all the other skin tumours were negative. On the other hand, almost all the seborrhoeic keratoses showed negative immunoreactivity for antiphoshohistone H2AX (gamma-H2AX) as a marker of OIS, but 17 of 22 cases (77%) of AK were moderately to strongly positive. Immunoreactivity for gamma-H2AX was significantly greater in AK than in seborrhoeic keratosis, BD, BCC and SCC. The activation of FGFR3 might be a common feature in the tumorigenesis in seborrhoeic keratosis, although the activation does not induce a typical oncogenic signal in keratinocytes. In addition, OIS due to some oncogenic signals rather than activation of FGFR3 might be involved in the early skin carcinogenesis related to chronic ultraviolet radiation exposure.

The rapid rise in skin diseases over the past decade has been a growing concern worldwide. Early detection, correct categorization, and accurate identification can result in the successful treatment of melanoma, thereby decreasing the morbidity and mortality rate. Thus, there is a significant need for a system that is capable of identifying skin diseases and precisely classifying them. The proposed work aims to develop a multi class classification system using transfer learning-based convolutional neural networks (CNN). In particular, the proposed solution classifies the dermoscopic images to 8 different categories namely Melanoma (MEL), Basal Cell Carcinoma (BCC), Actinic Keratosis (AK), Benign Keratosis (BKL), Dermatofibroma (DF), Vascular lesions (VASC) and Squamous Cell Carcinoma (SCC). Four state-of-art pre-trained models are used for this task. A functional model-based network is leveraged to embed these sub-models in a larger multi-headed neural network. This will allow the embedded model to be treated as a single large model. An ensemble approach, termed as blending, is employed to combine the predictions efficiently made by the sub-models. Additionally, a robust cropping strategy is implemented to deal with the uncropped images and their impact on the performance of the classifiers is investigated. The impact of applying blending technique to ensemble the pre-trained CNNs are investigated against the performance of the individual classifier. The proposed work is carried out on International Skin Imaging Collaboration (ISIC) 2019 dataset. In this work, the solution for task 1 of the challenge is presented and we obtained balanced multi-class accuracy of 81.2% on the dataset compiled from the original dataset.

Non-melanoma skin cancer is the most common malignancy that shows increasing incidence due to our cumulative exposure to ultraviolet irradiation. Its major subtypes, basal cell carcinoma (BCC) and squamous cell carcinoma (SCC) differ in pathobiology, phenotype and clinical behavior, which must be reflected at the molecular level. In this study, protein expression profiles of BCC and SCC were tested in tissue microarrays and correlated with that of actinic keratosis, Bowen's disease, seborrheic keratosis and normal epidermis by detecting 22 proteins involved in cell interactions, growth, cell cycle regulation or apoptosis. The significantly more reduced collagen XVII, CD44v6, pan-Desmoglein levels and more evident E-Cadherin delocalization in BCC compared to SCC correlated with the de novo dermal invasion of BCC against the progressive invasion from in situ lesions in SCC development. EGFR was also expressed at a significantly higher level in SCC than in BCC. The upregulated cell communication protein connexin43 in BCC could contribute to the protection of BCC from metastatic invasion. Elevated cell replication in BCC was underlined by the increased topoisomerase IIα and reduced p21(waf1) and p27(kip1) positive cells fractions compared to SCC. Compared to differentiated keratinocytes, caspase-8 and -9 were equally upregulated in skin carcinoma subtypes for either mediating apoptosis induction or immune escape of tumor cells. Hierarchical cluster analysis grouped SCC and actinic keratosis cases exclusively together in support of their common origin and malignant phenotype. BCC cases were also clustered fully together. Differentially expressed proteins reflect the distinct pathobiology of skin carcinoma subtypes and can serve as surrogate markers in doubtful cases.

Skin cancer is the most common cancer in the Western world. In order to accurately detect the disease, especially malignant melanoma-the most fatal form of skin cancer-at an early stage when the prognosis is excellent, there is an urgent need to develop noninvasive early detection methods. We believe that polarization speckle patterns, defined as a spatial distribution of depolarization ratio of traditional speckle patterns, can be an important tool for skin cancer detection. To demonstrate our technique, we conduct a large in vivo clinical study of 214 skin lesions, and show that statistical moments of the polarization speckle pattern could differentiate different types of skin lesions, including three common types of skin cancers, malignant melanoma, squamous cell carcinoma, basal cell carcinoma, and two benign lesions, melanocytic nevus and seborrheic keratoses. In particular, the fourth order moment achieves better or similar sensitivity and specificity than many well-known and accepted optical techniques used to differentiate melanoma and seborrheic keratosis.