A carregar...
Publicação
Detecting the Onset of Neurological Critical Events via Structure-Informed AI: Clinical Applications in Epilepsy and Amyotrophic Lateral Sclerosis
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 11.68 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Neurological disorders such as Epilepsy and Amyotrophic Lateral Sclerosis (ALS) pose profound
challenges due to their complex progression and significant impact on patients' quality of life.
Accurate and early detection is essential but remains difficult due to the variability and
subtlety of physiological signals. While machine learning and deep learning techniques have
shown potential in analyzing these signals, their clinical utility is often hindered by poor
interpretability, patient-specific variability, and class imbalance. This research addresses a
critical gap in the field: the limited application of causal AI to enhance the robustness,
generalization, and explainability of models used in neurological crisis detection. To this end,
two novel feature engineering strategies are introduced, each designed to capture aspects of
physiological data that traditional features often fail to represent. For seizure detection, a oneparameter family of features based on the powers of the covariance matrix is proposed,
offering a unique representation that captures latent causal structures governing emergent
macroscopic events. These features go beyond correlation and enable earlier and more
reliable detection of seizures. Indeed, we demonstrate that these structure-informed features
tend to be consistently separable in their natural ambient space, namely, the manifold of
Symmetric Positive Definite (SPD) matrices endowed with the Affine-Invariant Riemannian
metric, which both contributes to and helps explain the state-of-the-art classification
performance. For ALS gesture recognition, a distinct set of features derived from lag
correlation vectors is introduced, capturing the temporal dependencies and internal rhythmic
patterns of electromyography (EMG)signals through windowed lag-based comparisons. These
representations preserve signal dynamics often lost in conventional feature engineering,
providing a richer and more discriminative input for classification models. Together, these
innovations advance the application of causal inference and structure-informed signal
representations in clinical AI, offering new pathways toward interpretable and effective
diagnostic tools.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Information Management, specialization in Business Intelligence
Palavras-chave
Seizure Detection Gesture Recognition Brain-Computer Interface Causal AI Powers of the Covariance Matrices Lag Correlation Vectors SDG 3 - Good health and well-being SDG 4 - Quality education SDG 9 - Industry, innovation and infrastructure SDG 10 - Reduced inequalities