Incidence and profile of skin cancers in patients following ultraviolet phototherapy without psoralens: A retrospective cohort study.

BACKGROUND: Psoralen + ultraviolet-A (PUVA) is associated with photocarcinogenesis. However, carcinogenic risk with other ultraviolet phototherapies remains unclear. OBJECTIVE: Evaluate whether phototherapy without psoralens increases skin cancer risk. METHODS: Retrospective cohort study of patients treated at a teaching-hospital phototherapy center (1977-2018). Skin cancer records were validated against pathology reports. Age-standardized incidence rates (ASIRs) of skin cancer were evaluated for gender, skin phototype, diagnosis, ultraviolet modality, anatomical site; and compared to provincial population incidence rates (2003). RESULTS: In total, 3506 patients treated with broadband-ultraviolet-B, narrowband-UVB and/or combined UVAB were assessed with a mean follow-up of 7.3 years. Majority of patients had psoriasis (60.9%) or eczema (26.4%). Median number of treatments was 43 (1-3598). Overall, 170 skin cancers (17 melanoma, 33 squamous cell carcinoma and 120 basal cell carcinoma) occurred in 79 patients. Patient-based and tumor-based ASIR of skin cancer was 149 (95% CI: 112-187)/100,000 and 264 (219-309)/100,000 person-years, respectively. There was no significant difference between tumor-based ASIRs for melanoma, squamous cell carcinoma, and basal cell carcinoma compared to the general population; or in phototherapy patients with-psoriasis or eczema; or immunosuppressants. No cumulative dose-response correlation between UVB and skin cancer was seen. LIMITATIONS: Treatment and follow-up duration. CONCLUSION: No increased risk of melanoma and keratinocyte cancer was found with phototherapy.

Micro-relief characterization of benign and malignant skin lesions by polarization speckle analysis in vivo.

BACKGROUND/PURPOSE: A recent direction in skin disease classification is to develop quantitative diagnostic techniques. Skin relief, colloquially known as roughness, is an important clinical feature. The aim of this study is to demonstrate a novel polarization speckle technique to quantitatively measure roughness on skin lesions in vivo. We then calculate the average roughness of different types of skin lesions to determine the extent to which polarization speckle roughness measurements can be used to identify skin cancer. METHODS: The experimental conditions were set to target the fine relief structure on the order of ten microns within a small field of view of 3 mm. The device was tested in a clinical study on patients with malignant and benign skin lesions that resemble cancer. The cancer group includes 37 malignant melanomas (MM), 43 basal cell carcinomas (BCC), and 26 squamous cell carcinomas (SCC), all categories confirmed by gold standard biopsy. The benign group includes 109 seborrheic keratoses (SK), 79 nevi, and 11 actinic keratoses (AK). Normal skin roughness was obtained for the same patients (301 different body sites proximal to the lesion). RESULTS: The average root mean squared (rms) roughness ± standard error of the mean for MM and nevus was equal to 19 ± 5 µm and 21 ± 3 µm, respectively. Normal skin has rms roughness of 31 ± 3 µm, other lesions have roughness of 35 ± 10 µm (AK), 35 ± 7 µm (SCC), 31 ± 4 µm (SK), and 30 ± 5 µm (BCC). CONCLUSION: An independent-samples Kruskal-Wallis test indicates that MM and nevus can be separated from each of the tested types of lesions, except each other. These results quantify clinical knowledge of lesion roughness and could be useful for optical cancer detection.

InterMEL: An international biorepository and clinical database to uncover predictors of survival in early-stage melanoma.

INTRODUCTION: We are conducting a multicenter study to identify classifiers predictive of disease-specific survival in patients with primary melanomas. Here we delineate the unique aspects, challenges, and best practices for optimizing a study of generally small-sized pigmented tumor samples including primary melanomas of at least 1.05mm from AJTCC TNM stage IIA-IIID patients. We also evaluated tissue-derived predictors of extracted nucleic acids' quality and success in downstream testing. This ongoing study will target 1,000 melanomas within the international InterMEL consortium. METHODS: Following a pre-established protocol, participating centers ship formalin-fixed paraffin embedded (FFPE) tissue sections to Memorial Sloan Kettering Cancer Center for the centralized handling, dermatopathology review and histology-guided coextraction of RNA and DNA. Samples are distributed for evaluation of somatic mutations using next gen sequencing (NGS) with the MSK-IMPACTTM assay, methylation-profiling (Infinium MethylationEPIC arrays), and miRNA expression (Nanostring nCounter Human v3 miRNA Expression Assay). RESULTS: Sufficient material was obtained for screening of miRNA expression in 683/685 (99%) eligible melanomas, methylation in 467 (68%), and somatic mutations in 560 (82%). In 446/685 (65%) cases, aliquots of RNA/DNA were sufficient for testing with all three platforms. Among samples evaluated by the time of this analysis, the mean NGS coverage was 249x, 59 (18.6%) samples had coverage below 100x, and 41/414 (10%) failed methylation QC due to low intensity probes or insufficient Meta-Mixed Interquartile (BMIQ)- and single sample (ss)- Noob normalizations. Six of 683 RNAs (1%) failed Nanostring QC due to the low proportion of probes above the minimum threshold. Age of the FFPE tissue blocks (p<0.001) and time elapsed from sectioning to co-extraction (p = 0.002) were associated with methylation screening failures. Melanin reduced the ability to amplify fragments of 200bp or greater (absent/lightly pigmented vs heavily pigmented, p<0.003). Conversely, heavily pigmented tumors rendered greater amounts of RNA (p<0.001), and of RNA above 200 nucleotides (p<0.001). CONCLUSION: Our experience with many archival tissues demonstrates that with careful management of tissue processing and quality control it is possible to conduct multi-omic studies in a complex multi-institutional setting for investigations involving minute quantities of FFPE tumors, as in studies of early-stage melanoma. The study describes, for the first time, the optimal strategy for obtaining archival and limited tumor tissue, the characteristics of the nucleic acids co-extracted from a unique cell lysate, and success rate in downstream applications. In addition, our findings provide an estimate of the anticipated attrition that will guide other large multicenter research and consortia.

SSD-KD: A self-supervised diverse knowledge distillation method for lightweight skin lesion classification using dermoscopic images.

Skin cancer is one of the most common types of malignancy, affecting a large population and causing a heavy economic burden worldwide. Over the last few years, computer-aided diagnosis has been rapidly developed and make great progress in healthcare and medical practices due to the advances in artificial intelligence, particularly with the adoption of convolutional neural networks. However, most studies in skin cancer detection keep pursuing high prediction accuracies without considering the limitation of computing resources on portable devices. In this case, the knowledge distillation (KD) method has been proven as an efficient tool to help improve the adaptability of lightweight models under limited resources, meanwhile keeping a high-level representation capability. To bridge the gap, this study specifically proposes a novel method, termed SSD-KD, that unifies diverse knowledge into a generic KD framework for skin disease classification. Our method models an intra-instance relational feature representation and integrates it with existing KD research. A dual relational knowledge distillation architecture is self-supervised trained while the weighted softened outputs are also exploited to enable the student model to capture richer knowledge from the teacher model. To demonstrate the effectiveness of our method, we conduct experiments on ISIC 2019, a large-scale open-accessed benchmark of skin diseases dermoscopic images. Experiments show that our distilled MobileNetV2 can achieve an accuracy as high as 85% for the classification tasks of 8 different skin diseases with minimal parameters and computing requirements. Ablation studies confirm the effectiveness of our intra- and inter-instance relational knowledge integration strategy. Compared with state-of-the-art knowledge distillation techniques, the proposed method demonstrates improved performance. To the best of our knowledge, this is the first deep knowledge distillation application for multi-disease classification on the large-scale dermoscopy database. Our codes and models are available at https://github.com/enkiwang/Portable-Skin-Lesion-Diagnosis.

Landscape of mutations in early stage primary cutaneous melanoma: An InterMEL study.

It is unclear why some melanomas aggressively metastasize while others remain indolent. Available studies employing multi-omic profiling of melanomas are based on large primary or metastatic tumors. We examine the genomic landscape of early-stage melanomas diagnosed prior to the modern era of immunological treatments. Untreated cases with Stage II/III cutaneous melanoma were identified from institutions throughout the United States, Australia and Spain. FFPE tumor sections were profiled for mutation, methylation and microRNAs. Preliminary results from mutation profiling and clinical pathologic correlates show the distribution of four driver mutation sub-types: 31% BRAF; 18% NRAS; 21% NF1; 26% Triple Wild Type. BRAF mutant tumors had younger age at diagnosis, more associated nevi, more tumor infiltrating lymphocytes, and fewer thick tumors although at generally more advanced stage. NF1 mutant tumors were frequent on the head/neck in older patients with severe solar elastosis, thicker tumors but in earlier stages. Triple Wild Type tumors were predominantly male, frequently on the leg, with more perineural invasion. Mutations in TERT, TP53, CDKN2A and ARID2 were observed often, with TP53 mutations occurring particularly frequently in the NF1 sub-type. The InterMEL study will provide the most extensive multi-omic profiling of early-stage melanoma to date. Initial results demonstrate a nuanced understanding of the mutational and clinicopathological landscape of these early-stage tumors.

Polarization memory rate as a metric to differentiate benign and malignant tissues.

Non-invasive optical methods for cancer diagnostics, such as microscopy, spectroscopy, and polarimetry, are rapidly advancing. In this respect, finding new and powerful optical metrics is an indispensable task. Here we introduce polarization memory rate (PMR) as a sensitive metric for optical cancer diagnostics. PMR characterizes the preservation of circularly polarized light relative to linearly polarized light as light propagates in a medium. We hypothesize that because of well-known indicators associated with the morphological changes of cancer cells, like an enlarged nucleus size and higher chromatin density, PMR should be greater for cancerous than for the non-cancerous tissues. A thorough literature review reveals how this difference arises from the anomalous depolarization behaviour of many biological tissues. In physical terms, though most biological tissue primarily exhibits Mie scattering, it typically exhibits Rayleigh depolarization. However, in cancerous tissue the Mie depolarization regime becomes more prominent than Rayleigh. Experimental evidence of this metric is found in a preliminary clinical study using a novel Stokes polarimetry probe. We conducted in vivo measurements of 20 benign, 28 malignant and 59 normal skin sites with a 660 nm laser diode. The median PMR values for cancer vs non-cancer are significantly higher for cancer which supports our hypothesis. The reported fundamental differences in depolarization may persist for other types of cancer and create a conceptual basis for further developments in polarimetry applications for cancer detection.

Persistent organic pollutants and risk of cutaneous malignant melanoma among women.

BACKGROUND: Despite the increasing trend of cutaneous malignant melanoma (CMM) incidence in Canada, especially among females, few risk factors other than ultraviolet radiation exposure, have been identified. AIM: We conducted a case-control study of 406 CMM cases and 181 controls to evaluate the potential impact of body burdens of various persistent organic pollutants on CMM risk. METHODS: Detailed data on potential confounding factors, including lifetime repeated sun exposure and skin reaction to repeated sun exposure, were collected. Gas chromatography tandem mass spectrometry was used to assay plasma levels of 14 polychlorinated biphenyl (PCB) congeners and 11 organochlorine (OC) pesticides among cases and controls. RESULTS: Statistically significant trends of increased CMM risk were observed with increasing plasma concentrations of multiple PCB congeners, including PCBs 138, 153, 170, 180, 183 and 187. For example, compared to lowest plasma concentration quartile of PCB-138, the second, third and fourth quartiles were associated with 1.7 (95% CI: 0.9-2.9), 2.3 (95% CI: 1.3-4.1) and 2.4 (95% CI: 1.3-4.5) -fold increased risks of CMM, respectively. Similarly, increasing plasma concentrations of several OC pesticides (i.e., ß-HCH, HCB, Mirex, oxychlordane and trans-Nonachlor) showed statistically significant trends with increased CMM risk. For example, compared to lowest plasma concentration quartile of ß-HCH, the second, third and fourth quartiles were associated with 1.3 (95% CI: 0.7-2.3), 2.1 (95% CI: 1.2-3.7) and 2.3 (95% CI: 1.2-4.4) -fold increased risks of CMM, respectively. CONCLUSION: Plasma levels of several persistent organic pollutants were highly correlated, suggesting that observed associations were not necessarily independent of each other. Given the highly correlated nature of exposure to PCB and OC analytes, sophisticated analyses that consider complex mixtures should be considered in future studies.

Investigating the depolarization property of skin tissue by degree of polarization uniformity contrast using polarization-sensitive optical coherence tomography.

The depolarization property of skin has been found to be important for skin cancer detection. Previous techniques based on light polarization lack the capability of depth differentiation. Polarization-sensitive optical coherence tomography (PS-OCT) has the advantage of both depth-resolved 3D imaging and high sensitivity to polarization. In this study, we investigate the depolarization property of skin tissue using PS-OCT, especially with the degree of polarization uniformity (DOPU) contrast. Well designed skin phantoms with various surface roughness levels and optical properties mimicking skin are imaged by PS-OCT and the DOPU values are quantified. The result shows a correlation between DOPU and surface roughness, where a higher roughness corresponds to a lower DOPU value. An index matching experiment with a water layer confirms the impact of surface condition on light depolarization. Refraction of backscattered photons on the surface boundary is attributed to the broadening of backscattering angle and thus depolarization. To the best of our knowledge, this is the first time the impact of surface roughness on DOPU is reported and its mechanism explained. Furthermore, through preliminary in vivo skin imaging, the capability of DOPU in detecting depolarization in skin is demonstrated. By utilizing the 3D imaging from PS-OCT, DOPU can offer a high-resolution depth differentiation and quantification of depolarization in skin tissue.

Incorporating clinical knowledge with constrained classifier chain into a multimodal deep network for melanoma detection.

In recent years, vast developments in Computer-Aided Diagnosis (CAD) for skin diseases have generated much interest from clinicians and other eventual end-users of this technology. Introducing clinical domain knowledge to these machine learning strategies can help dispel the black box nature of these tools, strengthening clinician trust. Clinical domain knowledge also provides new information channels which can improve CAD diagnostic performance. In this paper, we propose a novel framework for malignant melanoma (MM) detection by fusing clinical images and dermoscopic images. The proposed method combines a multi-labeled deep feature extractor and clinically constrained classifier chain (CC). This allows the 7-point checklist, a clinician diagnostic algorithm, to be included in the decision level while maintaining the clinical importance of the major and minor criteria in the checklist. Our proposed framework achieved an average accuracy of 81.3% for detecting all criteria and melanoma when testing on a publicly available 7-point checklist dataset. This is the highest reported results, outperforming state-of-the-art methods in the literature by 6.4% or more. Analyses also show that the proposed system surpasses the single modality system of using either clinical images or dermoscopic images alone and the systems without adopting the approach of multi-label and clinically constrained classifier chain. Our carefully designed system demonstrates a substantial improvement over melanoma detection. By keeping the familiar major and minor criteria of the 7-point checklist and their corresponding weights, the proposed system may be more accepted by physicians as a human-interpretable CAD tool for automated melanoma detection.

Degree of optical polarization as a tool for detecting melanoma: proof of principle.

Determining the optical polarization properties of a skin lesion is a proposed method to differentiate melanoma from other skin lesions. We developed an in vivo Stokes polarimetry probe that fires a laser of known polarization at the skin and measures the Stokes parameters of the backscattered light in one shot. From these measured Stokes parameters, we can calculate the degree of polarization (DOP). Through testing on rough skin phantoms, a correlation between backscattered DOP and skin roughness was identified for both linear and circular input polarization, the latter of which was found to be more useful. In a pilot clinical trial of 69 skin lesions in vivo, it was found that the mean DOP for melanoma (linear input on melanoma: 0.46 ± 0.09) was greater than that of other lesions (linear input on all other lesions: 0.28 ± 0.01). This separation is greater for circular polarized input light, and it is likely that circular polarized light's greater sensitivity to surface roughness contributes to this result. In addition, all skin lesions demonstrated a stronger depolarizing effect on circular polarized light than linear polarized light. We have identified DOP as a potentially useful measurement to identify melanoma among other types of skin lesions.

Automated Detection and Segmentation of Vascular Structures of Skin Lesions Seen in Dermoscopy, With an Application to Basal Cell Carcinoma Classification.

Blood vessels are important biomarkers in skin lesions both diagnostically and clinically. Detection and quantification of cutaneous blood vessels provide critical information toward lesion diagnosis and assessment. In this paper, a novel framework for detection and segmentation of cutaneous vasculature from dermoscopy images is presented and the further extracted vascular features are explored for skin cancer classification. Given a dermoscopy image, we segment vascular structures of the lesion by first decomposing the image using independent-component analysis into melanin and hemoglobin components. This eliminates the effect of pigmentation on the visibility of blood vessels. Using k-means clustering, the hemoglobin component is then clustered into normal, pigmented, and erythema regions. Shape filters are then applied to the erythema cluster at different scales. A vessel mask is generated as a result of global thresholding. The segmentation sensitivity and specificity of 90% and 86% were achieved on a set of 500 000 manually segmented pixels provided by an expert. To further demonstrate the superiority of the proposed method, based on the segmentation results, we defined and extracted vascular features toward lesion diagnosis in basal cell carcinoma (BCC). Among a dataset of 659 lesions (299 BCC and 360 non-BCC), a set of 12 vascular features are extracted from the final vessel images of the lesions and fed into a random forest classifier. When compared with a few other state-of-art methods, the proposed method achieves the best performance of 96.5% in terms of area under the curve (AUC) in differentiating BCC from benign lesions using only the extracted vascular features.

Melanoma Epidemiology and Prevention.

The epidemiology of melanoma is complex, and individual risk depends on sun exposure, host factors, and genetic factors, and in their interactions as well. Sun exposure can be classified as intermittent, chronic, or cumulative (overall) exposure, and each appears to have a different effect on type of melanoma. Other environmental factors, such as chemical exposures-either through occupation, atmosphere, or food-may increase risk for melanoma, and this area warrants further study. Host factors that are well known to be important are the numbers and types of nevi and the skin phenotype. Genetic factors are classified as high-penetrant genes, moderate-risk genes, or low-risk genetic polymorphisms. Subtypes of tumors, such as BRAF-mutated tumors, have different risk factors as well as different therapies. Prevention of melanoma has been attempted using various strategies in specific subpopulations, but to date optimal interventions to reduce incidence have not emerged.

Skin lesion tracking using structured graphical models.

An automatic pigmented skin lesions tracking system, which is important for early skin cancer detection, is proposed in this work. The input to the system is a pair of skin back images of the same subject captured at different times. The output is the correspondence (matching) between the detected lesions and the identification of newly appearing and disappearing ones. First, a set of anatomical landmarks are detected using a pictorial structure algorithm. The lesions that are located within the polygon defined by the landmarks are identified and their anatomical spatial contexts are encoded by the landmarks. Then, these lesions are matched by labeling an association graph using a tensor-based algorithm. A structured support vector machine is employed to learn all free parameters in the aforementioned steps. An adaptive learning approach (on-the-fly vs offline learning) is applied to set the parameters of the matching objective function using the estimated error of the detected landmarks. The effectiveness of the different steps in our framework is validated on 194 skin back images (97 pairs).

Development and validation of a melanoma risk score based on pooled data from 16 case-control studies.

BACKGROUND: We report the development of a cutaneous melanoma risk algorithm based upon seven factors; hair color, skin type, family history, freckling, nevus count, number of large nevi, and history of sunburn, intended to form the basis of a self-assessment Web tool for the general public. METHODS: Predicted odds of melanoma were estimated by analyzing a pooled dataset from 16 case-control studies using logistic random coefficients models. Risk categories were defined based on the distribution of the predicted odds in the controls from these studies. Imputation was used to estimate missing data in the pooled datasets. The 30th, 60th, and 90th centiles were used to distribute individuals into four risk groups for their age, sex, and geographic location. Cross-validation was used to test the robustness of the thresholds for each group by leaving out each study one by one. Performance of the model was assessed in an independent UK case-control study dataset. RESULTS: Cross-validation confirmed the robustness of the threshold estimates. Cases and controls were well discriminated in the independent dataset [area under the curve, 0.75; 95% confidence interval (CI), 0.73-0.78]. Twenty-nine percent of cases were in the highest risk group compared with 7% of controls, and 43% of controls were in the lowest risk group compared with 13% of cases. CONCLUSION: We have identified a composite score representing an estimate of relative risk and successfully validated this score in an independent dataset. IMPACT: This score may be a useful tool to inform members of the public about their melanoma risk.

Hair enhancement in dermoscopic images using dual-channel quaternion tubularness filters and MRF-based multilabel optimization.

Hair occlusion is one of the main challenges facing automatic lesion segmentation and feature extraction for skin cancer applications. We propose a novel method for simultaneously enhancing both light and dark hairs with variable widths, from dermoscopic images, without the prior knowledge of the hair color. We measure hair tubularness using a quaternion color curvature filter. We extract optimal hair features (tubularness, scale, and orientation) using Markov random field theory and multilabel optimization. We also develop a novel dual-channel matched filter to enhance hair pixels in the dermoscopic images while suppressing irrelevant skin pixels. We evaluate the hair enhancement capabilities of our method on hair-occluded images generated via our new hair simulation algorithm. Since hair enhancement is an intermediate step in a computer-aided diagnosis system for analyzing dermoscopic images, we validate our method and compare it to other methods by studying its effect on: 1) hair segmentation accuracy; 2) image inpainting quality; and 3) image classification accuracy. The validation results on 40 real clinical dermoscopic images and 94 synthetic data demonstrate that our approach outperforms competing hair enhancement methods.

Spatial normalization of human back images for dermatological studies.

A large number of pigmented skin lesions (PSLs) are a strong predictor of malignant melanoma. Many dermatologists advocate total body photography for high-risk patients because detecting new-appearing, disappearing, and changing PSL is important for early detection of the disease. However, manual inspection and matching of PSL is a subjective, tedious, and error-prone task. A computer program for tracking the corresponding PSL will greatly improve the matching process. In this paper, we describe the construction of the first human back template (atlas), which is used to facilitate spatial normalization of the PSL during the matching process. Four pairs of anatomically meaningful landmarks (neck, shoulder, armpit, and hip points) are used as reference points on the back image. Using the landmarks, a grid with longitudes and latitudes is constructed and overlaid on each subject-specific back image. To perform spatial normalization, the grid is registered into the back template, a unit-square rectilinear grid. To demonstrate the benefits of using the back template, we apply several state-of-the-art point-matching algorithms on 56 pairs of real dermatological images and show that utilizing spatially normalized coordinates improves the PSL matching accuracies.

Extracutaneous melanoma epidemiology in British Columbia.

The epidemiology of extracutaneous melanoma (ECM) is sparsely reported upon in the literature, and studies to date have been limited both by time and by geographic gaps in available data. Utilizing a comprehensive provincial cancer registry, we sought to analyze the incidence and survival rates of ECM on the basis of sex and anatomic distribution for the British Columbia, Canada population. Data on ECMs diagnosed between 1 January 1992 and 31 December 2006 were obtained from the BC Cancer Registry. Anatomical sites of ECM were classified on the basis of ICD-9 codes, and incidence rates for each site were age standardized and grouped by sex. The 5-year survival rate for each anatomical grouping was tracked until 31 December 2011. A total of 922 primary ECMs were recorded in the BC Cancer Registry between 1992 and 2006, representing 5.1% of melanoma incidence. Ocular melanomas were most frequently reported, with an age-standardized incidence rate (per million) of 10.6 for men and 8.5 for women. ECM patients were generally older at diagnosis and had poorer survival rates compared with cutaneous melanoma cases. Five-year survival rates for ECM varied markedly from 23.5% for genital lesions to 87.0% for ocular cases. Our ECM epidemiology results are largely consistent with previous studies from the USA and Europe. Where considerable differences in reported values do exist, the opportunity arises to assess the efficacy of melanoma detection, monitoring, and treatment strategies in different geographic regions. Our study represents the largest epidemiological investigation of ECM in Canada and provides a framework for future epidemiological comparisons.

Detection and analysis of irregular streaks in dermoscopic images of skin lesions.

Irregular streaks are important clues for Melanoma (a potentially fatal form of skin cancer) diagnosis using dermoscopy images. This paper extends our previous algorithm to identify the absence or presence of streaks in a skin lesions, by further analyzing the appearance of detected streak lines, and performing a three-way classification for streaks, Absent, Regular, and Irregular, in a pigmented skin lesion. In addition, the directional pattern of detected lines is analyzed to extract their orientation features in order to detect the underlying pattern. The method uses a graphical representation to model the geometric pattern of valid streaks and the distribution and coverage of the structure. Using these proposed features of the valid streaks along with the color and texture features of the entire lesion, an accuracy of 76.1% and weighted average area under ROC curve (AUC) of 85% is achieved for classifying dermoscopy images into streaks Absent, Regular, or Irregular on 945 images compiled from atlases and the internet without any exclusion criteria. This challenging dataset is the largest validation dataset for streaks detection and classification published to date. The data set has also been applied to the two-class sub-problems of Absent/Present classification (accuracy of 78.3% with AUC of 83.2%) and to Regular/Irregular classification (accuracy 83.6% with AUC of 88.9%). When the method was tested on a cleaned subset of 300 images randomly selected from the 945 images, the AUC increased to 91.8%, 93.2% and 90.9% for the Absent/Regular/Irregular, Absent/Present, and Regular/Irregular problems, respectively.