Search

We present a new technique for melanocytic lesion segmentation, Mimicking Expert Dermatologists' Segmentations (MEDS), and extensive tests of its accuracy, speed, and robustness. MEDS combines a thresholding scheme reproducing the cognitive process of dermatologists with a number of optimizations that may be of independent interest. MEDS is simple, with a single parameter tuning its “tightness”. It is extremely fast, segmenting medium-resolution images in a fraction of a second even with the modest computational resources of a cell phone-an improvement of an order of magnitude or more over state-of-the-art techniques. And it is extremely accurate: very experienced dermatologists disagree with its segmentations less than they disagree with the segmentations of state-of-the-art techniques, and in fact less than they disagree with the segmentations of dermatologists of moderate experience.

Detection of new or rapidly evolving melanocytic lesions is crucial for early diagnosis and treatment of melanoma. We propose a fully automated pre-screening system for detecting new lesions or changes in existing ones, on the order of 2 - 3mm, over almost the entire body surface. Our solution is based on a multi-camera 3D stereo system. The system captures 3D textured scans of a subject at different times and then brings these scans into correspondence by aligning them with a learned, parametric, non-rigid 3D body model. This means that captured skin textures are in accurate alignment across scans, facilitating the detection of new or changing lesions. The integration of lesion segmentation with a deformable 3D body model is a key contribution that makes our approach robust to changes in illumination and subject pose.

Background: The first step in the analysis of a dermatoscopically imaged melanocytic lesion is segmentation--informally, isolating those points in the image belonging to the lesion from those belonging to the surrounding non-lesional skin. Although typically studied in the context of automated analysis, segmentation is a necessary step even for human operators who plan to evaluate quantitative features of a lesion (such as diameter or asymmetry)., Methods: In a double blind evaluation of the segmentation of 77 digital dermatoscopic images, we observed a significant inter-operator variability., Results: The area of the disagreement region was on average 15.28% of the area of the lesion itself, and in 10% of the cases it was more than 28%. More experienced dermatologists showed greater agreement among themselves than with less experienced dermatologists, and a slight tendency toward 'tighter' segmentations., Conclusion: The evaluation methodology addresses a number of crucial difficulties encountered in previous studies and may be of independent interest. Our results underscore the necessity of taking into account inter-operator variability in large epidemiological studies, in particular those involving less experienced dermatologists, and of striving toward techniques allowing greater standardization and replicability in the evaluation of the fundamental visual parameters of lesions., (© 2011 John Wiley & Sons A/S.)

We present a novel approach to the segmentation of psoriasis lesions in "full body" digital photographs potentially involving dozens or even hundreds of separate lesions. Our algorithm first isolates a set of zones that certainly correspond to lesional plaques based on chromatic information, and then expands these zones to achieve an accurate segmentation of plaques through a Geometric Active Contours method. The variability in segmentation between our algorithm and different human operators appears comparable to the variability between human operators.

VirtualShave is a novel tool to remove hair from digital dermatoscopic images. First, individual hairs are identified using a top-hat filter followed by morphological postprocessing. Then, they are replaced through PDE-based inpainting with an estimate of the underlying occluded skin. VirtualShave's performance is comparable to that of a human operator removing hair manually, and the resulting images are almost indistinguishable from those of hair-free skin.

In a double blind evaluation of 60 digital dermatoscopic images by 4 "junior", 4 "senior" and 4 "expert" dermatologists (dermatoscopy training respectively less than 1 year, between 1 and 5 years, and more than 5 years), a significant inter-operator variability was observed in melanocytic lesion border identification (with a disagreement of the order of 10 - 20% of the area of the lesions). Expert dermatologists showed greater agreement among themselves than with senior and junior dermatologists, and a slight tendency towards "tighter" segmentations. The human inter-operator variability was then used to evaluate the segmentation accuracy of 4 algorithms, representative of the 3 fundamental state-of-the-art automated segmentation techniques and of a fourth, novel, technique. Our evaluation methodology addresses a number of crucial difficulties encountered in previous studies and may be of independent interest. 3 of the 4 algorithms showed considerably less agreement with expert dermatologists than even senior and junior dermatologists did (with a disagreement of the order of 30% of the area of the lesions); the remaining algorithm, however, showed agreement with expert dermatologists comparable to that of other expert dermatologists.