In diagnosing skin melanoma, we suggest a new approach to defining one of the essential Dermoscopic criteria: the blue and white structure. Using just picture labels to show whether the function is there or not, we accomplish this purpose in this article. To do this, each image is depicted as a non-overlapping area bag that can or cannot be classified as a BWS instance in each field. MIL is learned to predict the labels for the bag picture in a probabilistic graphic model. As a result, in the image, we expect the classification of the BWS in any photo and locate the attribute in the view. There are studies with high-tech performance, and BWS Detection provides the best competing methods in terms of its quality on a complex dataset. This paper increases the spectrum of modeling for computerized skin lesion image processing. It proposes, in particular, a system for the detection by weakly labeled data of dermoscopic local properties. © 2023 Elsevier B.V., All rights reserved.