A deep learning approach to detect blood vessels in basal cell carcinoma

PURPOSE

Blood vessels called telangiectasia are visible in skin lesions with the aid of dermoscopy. Telangiectasia are a pivotal identifying feature of basal cell carcinoma. These vessels appear thready, serpiginous, and may also appear arborizing, that is, wide vessels branch into successively thinner vessels. Due to these intricacies, their detection is not an easy task, neither with manual annotation nor with computerized techniques. In this study, we automate the segmentation of telangiectasia in dermoscopic images with a deep learning U-Net approach.

METHODS

We apply a combination of image processing techniques and a deep learning-based U-Net approach to detect telangiectasia in digital basal cell carcinoma skin cancer images. We compare loss functions and optimize the performance by using a combination loss function to manage class imbalance of skin versus vessel pixels.

RESULTS

We establish a baseline method for pixel-based telangiectasia detection in skin cancer lesion images. An analysis and comparison for human observer variability in annotation is also presented.

CONCLUSION

Our approach yields Jaccard score within the variation of human observers as it addresses a new aspect of the rapidly evolving field of deep learning: automatic identification of cancer-specific structures. Further application of DL techniques to detect dermoscopic structures and handle noisy labels is warranted.

Structured zinc oxide dressing for secondary intention wounds

OBJECTIVE

Inexpensive methods for more rapid healing of secondary intention wounds are sought. This pilot study measured the wound healing rate for a new zinc oxide structured dressing technique.

METHOD

In this study, we included the three patients with the largest wounds with onset during a one month period. A 3-ply gauze was cut and placed in the centre of each wound, leaving a 3-5mm rim of the wound exposed to the zinc gauze. The central gauze was soaked with 0.9% saline solution and the entire wound area was covered with 3 layers of Unna zinc oxide dressing. The central gauze size was modified to fit as the wound healed and the size changed. The wound was photographed at each visit and wound areas were obtained using best-fit ellipses.

RESULTS

The average wound closure rate is reported in the three wounds as 21.73mm² per day. The scalp wound healed at a rate of 21.45mm² per day.; the spider bite wound healed at a rate of 28.92mm² per day; and the thigh wound healed at a rate of 14.81mm² per day.

CONCLUSION

Healing rates for the zinc gauze method exceed those previously reported for full-thickness wounds healing by secondary intention. Additional study of the new technique with more patients is needed before conclusions relevant to clinical practice can be made.

Biologically inspired skin lesion segmentation using a geodesic active contour technique

BACKGROUND/PURPOSE

Computer-aided diagnosis of skin cancer requires accurate lesion segmentation, which must overcome noise such as hair, skin color variations, and ambient light variability.

METHODS

A biologically inspired geodesic active contour (GAC) technique is used for lesion segmentation. The algorithm presented here employs automatic contour initialization close to the actual lesion boundary, overcoming the 'sticking' at minimum local energy spots caused by noise artifacts such as hair. The border is significantly smoothed to mimic natural lesions. In addition, features that mimic biological parameters include spectral image subtraction and removal of peninsulas and inlets. Multiple boundary choices borders are created by parameter options used at different steps. These choices can allow future improvement over the basic default border.

RESULTS

The basic GAC algorithm was tested on 100 images (30 melanomas and 70 benign lesions), yielding a median XOR border error of 6.7%, comparable to the median inter-dermatologist XOR border error (7.4%), and lower than the gradient vector flow snake median XOR error of 14.2% on the same image set. On a difficult low-contrast border set of 1238 images, which included 350 non-melanocytic lesions, a median XOR error of 23.9% is obtained.

CONCLUSION

GAC techniques show promise in attaining the goal of automatic skin lesion segmentation.

Real-time supervised detection of pink areas in dermoscopic images of melanoma: importance of color shades, texture and location

BACKGROUND/PURPOSE

Early detection of malignant melanoma is an important public health challenge. In the USA, dermatologists are seeing more melanomas at an early stage, before classic melanoma features have become apparent. Pink color is a feature of these early melanomas. If rapid and accurate automatic detection of pink color in these melanomas could be accomplished, there could be significant public health benefits.

METHODS

Detection of three shades of pink (light pink, dark pink, and orange pink) was accomplished using color analysis techniques in five color planes (red, green, blue, hue, and saturation). Color shade analysis was performed using a logistic regression model trained with an image set of 60 dermoscopic images of melanoma that contained pink areas. Detected pink shade areas were further analyzed with regard to the location within the lesion, average color parameters over the detected areas, and histogram texture features.

RESULTS

Logistic regression analysis of a separate set of 128 melanomas and 128 benign images resulted in up to 87.9% accuracy in discriminating melanoma from benign lesions measured using area under the receiver operating characteristic curve. The accuracy in this model decreased when parameters for individual shades, texture, or shade location within the lesion were omitted.

CONCLUSION

Texture, color, and lesion location analysis applied to multiple shades of pink can assist in melanoma detection. When any of these three details: color location, shade analysis, or texture analysis were omitted from the model, accuracy in separating melanoma from benign lesions was lowered. Separation of colors into shades and further details that enhance the characterization of these color shades are needed for optimal discrimination of melanoma from benign lesions.

Discrimination of squamous cell carcinoma in situ from seborrheic keratosis by color analysis techniques requires information from scale, scale-crust and surrounding areas in dermoscopy images

Scale-crust, also termed "keratin crust", appears as yellowish-to-tan scale on the skin's surface. It is caused by hyperkeratosis and parakeratosis in inflamed areas of squamous cell carcinoma in situ (SCCIS, Bowen's disease) and is a critical dermoscopy feature for detecting this skin cancer. In contrast, scale appears as a white-to-ivory detaching layer of the skin, without crust, and is most commonly seen in benign lesions such as seborrheic keratoses (SK). Distinguishing scale-crust from ordinary scale in digital dermoscopy images holds promise for early SCCIS detection and differentiation from SK. Reported here are image analysis techniques that best characterize scale-crust in SCCIS and scale in SK, thereby allowing accurate separation of these two dermoscopic features. Classification using a logistic regression operating on color features extracted from these digital dermoscopy structures can reliably separate SCCIS from SK.

Seasonality of brown recluse populations is reflected by numbers of brown recluse envenomations

A significant seasonal correlation was recently shown for brown recluse spider activity. Vetter (2011) observed brown recluse spiders were submitted by the general public predominantly during April-October. For patients with suspected brown recluse spider bites (BRSB), we have observed the same seasonality. Among 45 cases with features consistent of a BRSB, 43 (95.6%) occurred during April-October. Both the Vetter study and our study serve to demonstrate seasonal activity for brown recluse spiders.

Cloudy and starry milia-like cysts: how well do they distinguish seborrheic keratoses from malignant melanomas?

BACKGROUND

Seborrheic keratoses are the most common skin lesions known to contain small white or yellow structures called milia-like cysts (MLCs). Varied appearances can sometimes make it difficult to differentiate benign lesions from malignant lesions such as melanoma, the deadliest form of skin cancer found in humans.

OBJECTIVE

The purpose of this study was to determine the statistical occurrence of MLCs in benign vs. malignant lesions.

METHODS

A medical student with 10 months experience in examining approximately 1000 dermoscopy images and a dermoscopy-naïve observer analysed contact non-polarized dermoscopy images of 221 malignant melanomas and 175 seborrheic keratoses for presence of MLCs.

RESULTS

The observers found two different types of MLCs present: large ones described as cloudy and smaller ones described as starry. Starry MLCs were found to be prevalent in both seborrheic keratoses and melanomas. Cloudy MLCs, however, were found to have 99.1% specificity for seborrheic keratoses among this group of seborrheic keratoses and melanomas.

CONCLUSION

Cloudy MLCs can be a useful tool for differentiating between seborrheic keratoses and melanomas.

Border detection on digitized skin tumor images

A radial search technique is presented for detecting skin tumor borders in clinical dermatology images. First, it includes two rounds of radial search based on the same tumor center. The first-round search is independent, and the second-round search is knowledge-based tracking. Then a rescan with a new center is used to solve the blind-spot problem. The algorithm is tested on model images with excellent performance, and on 300 real clinical images with a satisfactory result.

Neural network diagnosis of malignant melanoma from color images

Malignant melanoma is the deadliest form of all skin cancers. Approximately 32,000 new cases of malignant melanoma were diagnosed in 1991 in the United States, with approximately 80% of patients expected to survive five years [1]. Fortunately, if detected early, even malignant melanoma may be treated successfully. Thus, in recent years, there has been rising interest in the automated detection and diagnosis of skin cancer, particularly malignant melanoma [2]. In this paper, we present a novel neural network approach for the automated separation of melanoma from three benign categories of tumors which exhibit melanoma-like characteristics. Our approach uses discriminant features, based on tumor shape and relative tumor color, that are supplied to an artificial neural network for classification of tumor images as malignant or benign. With this approach, for reasonably balanced training/testing sets, we are able to obtain above 80% correct classification of the malignant and benign tumors on real skin tumor images.

Detection of skin tumor boundaries in color images

A simple and yet effective method for finding the borders of tumors is presented as an initial step towards the diagnosis of skin tumors from their color images. The method makes use of an adaptive color metric from the red, green, and blue planes that contains information for discriminating the tumor from the background. Using this suitable coordinate transformation, the image is segmented. The tumor portion is then extracted from the segmented image and borders are drawn. Experimental results that verify the effectiveness of this approach are given.

Automatic detection of asymmetry in skin tumors

Asymmetry, a critical feature in the diagnosis of malignant melanoma, is analyzed using a new algorithm to find a major axis of asymmetry and calculate the degree of asymmetry of the tumor outline. The algorithm provides a new objective definition of asymmetry. A dermatologist classified 86 tumors as symmetric or asymmetric. Borders of tumors were found either manually or automatically using a radial search method. With either method, asymmetry determination by the asymmetry algorithm agreed with the dermatologist's determination of asymmetry in about 93% of cases.

Automatic detection of irregular borders in melanoma and other skin tumors

An irregularity index previously developed is applied to detect irregular borders automatically in skin tumor images, particularly malignant melanoma. The irregularity index is used to classify various tumor borders as irregular or regular. This procedure processes tumor images with borders automatically determined by a radial search algorithm previously described. Potential use of this algorithm in an in vivo skin cancer detection system and errors expected in the use of the algorithm are discussed.

Texture in skin images: comparison of three methods to determine smoothness

Smooth texture, a critical feature in skin tumor diagnosis, is analyzed using three texture measurement methods. A dermatologist classified 1290 small blocks within 42 tumor images as smooth, partially smooth, or nonsmooth. Texture discriminatory power of three methods were compared: the neighboring gray-level dependence matrix (NGLDM) method of Sun and Wee, the circular symmetric autoregressive random field model of Kashyap and Khotanzad, and a new peak-variance method. The texture analysis method that allows best prediction of smoothness for our tumor domain is the NGLDM method, affording 98% correct prediction of a smooth block with 21% false positives. We discuss applicability of texture analysis to dermatology.

Digital imaging in dermatology

In this article we discuss the recent surge in activity in digital imaging in dermatology. The key role of digital imaging as an adjunct to detection of early malignant melanoma, with application in following patients with the dysplastic nevus syndrome, is explored. Other current and future uses of digital imaging in image archiving, in clinical studies such as hair growth studies, and in telediagnosis are reviewed. We review the varying research activities of image analysis laboratories participating in the dermatology image researching group. Research laboratories included in this group are at Oregon Health Sciences University, Xerox Corporation, University of Arizona, University of Cincinnati, University of Munich, University of Wurzburg, University of Arkansas, Harvard University, Southern Illinois University-Edwardsville, Johns Hopkins University, National Institutes of Health, and University of Missouri at Columbia and Rolla. The role of new imaging devices in dermatology including the "nevoscope" and the dermatoscope is explored. Goals and challenges for the new technology are discussed.

An automatic color segmentation algorithm with application to identification of skin tumor borders

A principal components transform algorithm for automatic color segmentation of images is described. This color segmentation algorithm was used to find tumor borders in six different color spaces including the original red, green, and blue (RGB) color space of the digitized image, the intensity/hue/saturation (IHS) transform, the spherical transform, chromaticity coordinates, the CIE transform and the uniform color transform designated CIE-LUV. Five hundred skin tumor images were separated into a training set and a test set for comparison of the different color spaces. Automatic induction was applied to dynamically determine the number of colors for segmentation. Ninety-one percent of image variance was contained in the image component along the principal axis (also containing the most image information). When compared to a luminance radial search method, the principal components color segmentation border method performed equally well by one measure and 10% better by another measure, including more near border points outside the tumor. The spherical transform provides the highest success rate and the chromaticity transform the lowest error rate, although large variances in the data preclude definitive statistical comparisons.

Skin cancer recognition by computer vision

Automatic detection of several features characteristic of basal cell epitheliomas is described. The features selected for this feasibility study are semitranslucency, telangiectasia, ulcer, crust, and tumor border. Image processing methods used in this study include frequency analysis of the Fourier transform of the image, the Sun-Wee texture analysis algorithm, and several other image analysis techniques suitable for skin photographs. This image analysis software is designed for use with AI/DERM, an expert system that models diagnosis of skin tumors by dermatologists.

Computer-aided diagnosis of dermatologic disorders

In conclusion, several computer programs have been developed to aid diagnosis in dermatology. These have used various methods to reach a diagnosis, with some earlier programs using probabilistic techniques and later programs using cognitive models. Several current trends, if they continue, would indicate that computer-assisted diagnosis may play a greater role. The growing number of physicians who own the prerequisite computing resources will probably want diagnostic software to accompany their office management and data base software. The increasing ease of use of diagnostic software, with rapid access and ease of updating, should make these programs more attractive to physicians. Some of the newer peripheral devices such as optical disks and voice input will probably make the software more interesting. Perhaps the most important factor governing the acceptance of computer diagnostic adjuncts is whether the software will provide a useful service to physicians. If these programs can be demonstrated to improve patient care, they may become commonplace in physicians' offices.