The proposed framework is designed to facilitate the diagnostic process for autism spectrum disorder by combining advanced machine learning strategies with transparency through explainable AI techniques. The first module employs a deep learning model that has been fine-tuned using cross-domain transfer learning, which is particularly relevant given the challenges posed by limited data in ASD research. By enhancing data representation and adaptability, this approach allows for improved classification accuracy of ASD cases.
The second module of the framework focuses on elucidating the model’s decision-making processes, which is critical for clinical acceptance. Employing three different explainable AI techniques—saliency mapping, Gradient-weighted Class Activation Mapping, and SHapley Additive exPlanations (SHAP) analysis—this work uncovers the brain regions most significantly associated with ASD diagnoses. By correlating the model’s findings with established neurobiological evidence, the study not only demonstrates the effectiveness of the diagnostic framework but also underscores its clinical relevance.
👉 Pročitaj original: arXiv AI Papers
