Enhanced Content-Based Medical Image Retrieval in Healthcare Decision Support System Using AutoML and Vector Space Model

Machine learning techniques can potentially revolutionise healthcare; training such models demands significant expertise. We aim to assess the effectiveness of automated machine learning modules in empowering healthcare professionals with limited coding skills and no deep learning expertise to develop Artificial intelligence-based radiologic diagnostic systems. Content-based image retrieval (CBIR) approaches are gaining traction in healthcare, facilitating medical image archives to support clinical decision-making, especially in the vivid research area of analysing Magnetic resonance imaging (MRI) scans for brain tumours. In this paper, we propose an innovative Contentbased medical image retrieval (CBMIR) pipeline that integrates Run-length matrix (RLM) and Discrete wavelet transform (DWT) features and employs an Automated machine learning (AutoML) clustering approach within the Vector space model (VSM) for image indexing. AutoML facilitates efficient clustering and improves the indexing of images for retrieval. The proposed pipeline was evaluated on the BraTS 2018 MRI dataset and achieved a retrieval accuracy of 99.0%