Automatic detection of hemangiomas using unsupervised segmentation of regions of interest

Automatic Segmentation and Measurement of Infantile Hemangioma

“…The RG-FCM proposed in [10] is an improved region growing algorithm based on FCM, and gives better results than that of FCM. The seeds used as starting points in region growing are computed from the means obtained by applying FCM combined with a threshold based on the minimum value of the hemangioma class and the maximum value of the non-hemangioma class.…”

“…The performances of three segmentation methods (Otsu, fuzzy C-means (FCM), and a region growing algorithm based on FCM, RG-FCM) were compared in [10]; the best result, 91.51%, was obtained with RG-FCM. In [11] the image was segmented in 25 classes using a self-organizing map (SOM) network, and then the number of classes was reduced to only two classes (hemangioma and non-hemangioma) with a morphological approach.…”

Fleckmentation: rapid segmentation using repeated 2‐means

“…Although the successful you only look once (YOLO) network appears to defy the need for any segmentation [32], it nevertheless might profit from a fast decomposition as presented here; perhaps even depth = 2 could be sufficient. No claim is made that this is a complete segmentation method, as we rather believe that ensembles of techniques are the solution to solve complex tasks perfectly (see also [33,34]).…”

Melanoma Recognition with an Ensemble of Techniques for Segmentation and a Structural Analysis for Classification

“…The algorithm is applied with a limited range of ks, i.e. [2,3,4,5,6,7,8]. For each clustering outcome, its regions are determined; a region consists of a set of contiguous pixels.…”

“…Recently there has been a shift to combining multiple techniques, because a single technique often fails in certain circumstances, for which another technique can perform well. For instance, Celebi et al use multiple single (global) threshold methods [6]; Neghina et al combine the method of region growing with a method of pixel clustering [7]. In this study, we combine multi-level (global) thresholding with a pixel-clustering method (Section II).…”