Explainable Fetal Ultrasound Classification with CNN and MLP Models

Artificial Intelligence has greatly influenced healthcare, most particularly in medical imaging. This paper represents a review in large form that classifies fetal ultrasound images with the use of convolutional neural networks and multi-Layer Perceptrons. While CNN is very good at spatial feature extraction in image classification, their lack of interpretability presents challenges toward applications in health. In this regard, we include methods of Explainable AI (XAI), more precisely Local Interpre table Model-Agnostic Explanations (LIME), for giving more transparency and confidence in the decision-making process of such models. The research here utilizes 12,400 fetal ultrasound images, which were classified under six anatomical structures. The CNN and MLP models showed very promising classification performances of 93.24% and 91.17%, respectively. LIME was implemented to interpret model predictions and to more clearly identify factors contributing to the classification. The results also show that explainability enhances not only trust in AI-based diagnostics but also model reliability in clinical settings.