Название: Artificial Intelligence: A tool for effective diagnostics
Автор: Smith K Khare, Sachin Taran, Ankush D Jamthikar
Издательство: IOP Publishing
Серия: IOP Series in Artificial Intelligence in the Biomedical Sciences
Год: 2024
Страниц: 339
Язык: английский
Формат: pdf (true), epub
Размер: 21.2 MB

The book explores the application of Artificial Intelligence (AI) across various human–machine interfaces, addressing areas such as human attention, emotions, seizures, Alzheimer’s disease, focal and non-focal disorders, electrocardiogram rhythms, abnormal heartbeats, and leukemia. It provides a thorough examination of techniques for analyzing and processing both physiological and physical signals, as well as smear blood images. Physiological signals discussed include electroencephalograms (EEGs), electrocardiograms (ECGs), and electronic health records (EHRs), while physical signals encompass human speech. Serving as a comprehensive guide, the book delves into advanced signal processing techniques and the use of Machine Learning and Deep Learning for automated signal pre-processing and classification. Part of IOP Series in Artificial Intelligence in the Biomedical Sciences.

Many AI models are ‘black boxes,’ with decision-making mechanisms that are hidden and difficult for humans to understand. The opaque nature of AI models raises concerns among clinical experts and stakeholders, restricting the clinical applications of these models. Explainable artificial intelligence (XAI) has been developed with the goal of making AI judgments more transparent and intelligible. XAI plays an important role for several reasons, including trust, accuracy, fairness, responsibility, compliance, and AI system improvement. XAI plays a crucial role in achieving the objectives discussed below. Traditional AI models, particularly DL models, sometimes function as ‘black boxes,’ making judgments without revealing information about how those decisions were reached. XAI simplifies these processes, allowing consumers to comprehend the reasoning behind AI judgments. This openness is critical for fostering confidence between AI systems and stakeholders, ensuring that AI is perceived as a trustworthy and accountable tool.

The human voice is an essential tool for conveying emotions; therefore, this chapter proposes a novel speech emotion recognition (SER) framework based on a combination of empirical wavelet transform (EWT) and cubic support vector machine (CSVM). The system is designed for multiclass SER and uses a combination of auditory spectrograms, cepstral coefficients, spectral descriptors, periodicity, and harmonicity for feature extraction. The extracted features are used as CSVM classifier inputs, and the classifier is trained to recognize the speaker’s emotional state. In recent years, various techniques have been proposed for SER, including Machine Learning algorithms, feature extraction methods, and signal processing techniques. The Machine Learning algorithms most frequently employed for SER include artificial neural networks (ANNs), the k-nearest neighbors algorithm, support vector machines (SVMs), and decision trees.