Designing a cognitive smart healthcare framework for seizure prediction using multimodal convolutional neural network

Abstract

The Internet of Things (IoT) has evolved from a network of embedded computer devices to a network of brainy sensors. With the advent of IoT-cloud technologies, however, there is a growing demand for a cognitive framework that can deliver affordable, high-quality smart healthcare with a focus on the discrete patient. The advent of AI and deep learning methods makes it possible to include human-level reasoning into intelligent healthcare infrastructure. To evaluate the healthcare system, we also present a deep learning approach for detecting (Electroencephalography) EEG seizures. Extracting features from produced modal data, fusing multimodal feature maps, and determining whether or not the EEG signals have seizure are the separate responsibilities of the Multimodal Convolutional Neural Network (MMFCNN)’s three blocks. In addition, the study suggests a unique attention method, i.e., the cross-attention mechanism, to realise the information interaction and emphasise more representative info during the feature extraction stage. The study records and transmits EEG signals from epileptic patients using smart EEG sensors (in addition to general healthcare smart sensors). The cognitive framework then makes a judgement in the here-and-now regarding future actions and whether or not to submit the data to the deep learning unit. The suggested system detects the patient’s condition based on the latter’s actions, gestures, and facial expressions. Using a likelihood score, signals are processed and sorted into seizure and non-seizure categories in the cloud, where seizures are also detected. Those involved in a patient’s care receive the data so that they can keep tabs on them and, in emergency situations, take decisive action. The suggested model achieves 99.2% accuracy in experiments, as shown by the results.