Hierarchical Classification of Cough Sound Using Fine and Coarse-Grain Acoustic Feature Representations: A Doctor's...

From a medical perspective, gender and age group are critical parameters that influence both the diagnosis and treatment of various health conditions. Several works have been carried out on disease prediction; however, identification of age group and gender has not yet been extensively explored. We introduce a Hierarchical Multi-Task Learning (H-MTL) model designed for the diagnostic classification of respiratory conditions using the COUGHVID dataset. The dataset, initially structured non-hierarchically, is re-categorized into a multi-level taxonomy. The re-categorization introduces hierarchical labels based on gender, age group, and respiratory conditions, enabling the model to extract both fine and coarse-grain acoustic feature representations. This structure not only facilitates nuanced learning but also aligns with medical diagnostic reasoning by progressively narrowing down potential diagnoses. The novelty of the H-MTL model that seamlessly incorporates hierarchical dependencies into the learning process emulates the decision-making process of healthcare professionals by leveraging a shared feature extraction backbone and task-specific branches that hierarchically integrate predictions. Comparative experiments with Multi-Task Learning (MTL) and Multi-Class Classification (MCC) models demonstrate that the H-MTL model outperforms alternatives in classifying gender, age group, and respiratory condition, achieving superior accuracy, recall, and F1 scores.Clinical relevance- The H-MTL model's ability to integrate gender, age, and respiratory condition information in a hierarchical framework offers clinicians a more precise tool for diagnosing respiratory illnesses. By mimicking the decision-making process of healthcare professionals, it provides a nuanced, patient-specific diagnostic approach that could enhance early detection and personalized treatment strategies.