Blockchain-enabled federated learning with capsule network and incremental extreme learning machines for gastrointestinal bleeding detection in wireless capsule endoscopy

In the medical field, a wireless capsule endoscopy (WCE) diagnose gastrointestinal (GI) bleeding without any irritation to the patients. Unfortunately, there are still some problems with the accuracy of WCE images due to the substantial number of data used. The main challenges for the researchers in detecting GI bleeding are the rapid growth of the infection and the lack of reliable testing. Artificial intelligence (AI) makes GI bleeding diagnosis easier. Data sharing between different hospitals while maintaining confidentiality is challenging. This study creates a AI method that gathers data from four hospitals and uses blockchain-based federated learning (FL) to build a proposed model. Blockchain technology (BCT) is utilized to check the accuracy and quality of data and FL is used to train the model globally while upholding the organizations confidentiality. Firstly, we developed a data normalization technique to handle data collected from four sources. Secondly, we classify GI bleeding patients by employing incremental extreme learning machines (IELMs) and capsule networks (CapsNet). Lastly, we present a technique for jointly developing a global model with BCT and FL to ensure privacy. The performance of the proposed model is compared with five deep learning (DL) models for predicting GI bleeding using detailed testing on stomach WCE images while securing the data privacy of various users. Our results shows an improvement in recognizing GI bleeding individuals, with a 98.23 % accuracy rate. Furthermore, this model enhance the skills of medical professionals in diagnosing GI bleeding resulting in better and more efficient medical decision-making procedures.