Medical social network content analysis for medical image retrieval purpose

Abstract

SummaryBackground and objective: Medical social networking platforms provide virtual spaces ensuring the interaction between different healthcare participants. As a part of the exchange, these spaces allow subscribers to upload medical images, describing different medical cases for an analysis or an interpretation proposal. Facing this expected huge amount of uploaded images generated daily, it is needed to engage new mechanisms to effectively deal with this circumstance, for enhancing the search function process of medical images, based on what is uploaded. To overcome this issue, setting up of images visual searching based on a content‐based medical image retrieval scheme is the solution. More clearly, such mechanism will help and motivate medical social networking subscribers to find visually similar stored images.Methods: To ensure this task, the development of this mechanism, technically, is based mainly on a fusion of three visual features, which offers a flexible and more precision. It is reinforced by a weighted distance approach through attributing weights for feature vectors to scale up the performance. Indeed, the displayed results of this system can be updated based on user's intention by a user interactive feedback mechanism to indicate the truly relevant images.Results: We provide the theoretical performance of our scheme. Extensive experiments were conducted on a categorically classified collection containing 500 images. We conduct a practical evaluation on this dataset classes, putting returned results in a comparative study with other models results, existing in the literature.Conclusions: The proposed scheme preserves the efficiency of the search task. As theoretically and experimentally established, our scheme offers an effective image retrieval model that can support different subscribers' expectations. The relevance feedback mechanism can keep the dynamism of the system, thus offering a continuous searching result evolution. Experimentation outcomes indicate better findings compared with the other models.