Adaptive transformer-based deep learning model for fetal abnormal health and status assessment using ultrasound imagery

Fetal development is a crucial stage in prenatal care, necessitating the prompt detection of abnormalities in ultrasound scans to ensure the well-being of both the fetus and the expectant mother. Medical imaging is vital for identifying fetal abnormalities. However, despite considerable advancements in ultrasound technology, accurately identifying abnormalities in prenatal images remains challenging, often requiring significant time and expertise from healthcare specialists. AI is transforming medical imaging for diagnosis, with deep learning (DL) leading image analysis due to its exceptional pattern recognition, specifically in domains like fetal imaging, where several studies demonstrate DL aids in detecting abnormalities and measurements, supporting clinicians beyond human capabilities by automating intricate tasks and offering faster, more precise insights. In this article, we concentrate on the development of an Adaptive Transformer-Based Deep Learning Model for Fetal Abnormal Health and Status Assessment (ATDL-FAHSA) using Ultrasound Imagery. The main objective of the ATDL-FAHSA approach is to accurately classify fetal health, demonstrating significant potential for clinical deployment to prevent maternal and child mortality, thereby contributing to improved prenatal care outcomes. Initially, an input images endure systematic pre-processing, involving noise reduction, intensity normalization, and contrast enhancement to improve the clarity and diagnostic quality of fetal anatomical structures. Subsequently, deep spatial features are extracted using the ConvNeXtV2 model, enabling the capture of fine-grained morphological patterns from ultrasound images. For the classification process, a hybrid deep learning architecture utilizing an attention-driven DL network is employed. The proposed ATDL-FAHSA architecture adaptively merges convolutional and transformer-based reasoning, resulting in robust performance for both abnormality detection. The simulation analysis of the ATDL-FAHSA model has been performed employing the Ultrasound Fetus Dataset from the Kaggle repository. Extensive comparative result reports the encouraging performance of the ATDL-FAHSA approach over recent state-of-the-art models.