AI-Enabled Multimodal Prenatal Assessment Platform Utilizing Ultrasound Imaging and Amniotic Fluid Analysis for Early Anomaly Detection

The possibility of early detection of embryonic abnormalities has been a challenge as it depends on the use of imaging of the embryo through the use of ultrasound, which is prone to the acquisition noise and variability among the operators. The research reveals a framework of multimodal prenatal assessment, which is a combination of anatomical features identified by ultrasound and systematic biochemical biomarkers of an amniotic fluid to identify defects during the early stages of development. A convolutional neural network is used to gain hierarchical spatial representations of ultrasound scans, a multilayer perceptron to encode physiological biomarker data. The fusing of the heterogeneous feature spaces is done by employing a probabilistic neural network which estimates the posterior class distributions and the probability of prediction error is explicitly captured. The developed framework was tested on 1, 200 prenatal cases through stratified training and testing procedures. It is shown that experimental outcomes exhibit an accuracy of 93.8, a sensitivity of 91.6, and a AUC of 0.96, which is better than unimodal baselines. Uncertainty-aware fusion with SHAP-based interpretability offers the clinically transparent risk assessment, which can be effective in business early prenatal decisions.