In prenatal diagnosis, exact measurement of gestational age (GA) and fetal head circumference (HC) is crucial. Conventional methods rely on labor-intensive manual annotations, which vary depending on inter-observer accuracy. This work proposes Swin-UNetR, a Transformer-based deep learning architecture for automatic HC segmentation and GA prediction leveraging large-scale prenatal ultrasonic datasets. On three benchmark datasets—the HC18 Challenge (1,334 images), the Large-Scale Fetal Head Biometry Dataset (3,832 images), and the Multi-Hospital Fetal Ultrasound Dataset (6-class anatomical planes)—proposed model achieves state-of-the-art performance using self-supervised pre-training, hierarchical feature aggregation, and deformable attention mechanisms. The model performs excellently, illustrated by a Dice Similarity Coefficient (DSC) of 98.3% for HC segmentation and a mean absolute error (MAE) of 2.4 ± 1.9 days for GA prediction. This is considerably higher than other U-Net and CNN-Transformation hybrids. Swin-UNetR, and this was developed for clinical use, may generate predictions in actual time at a rate of 45 ms per picture. The program utilizes Grad-CAM attention mapping to generate predictions that's relatively simple to apply. The result illustrates that Swin-UNetR is a dependable, effective, and scalable way to standardize fetal biometry. This improves prenatal ultrasound evaluations better by making them less dependent on the ultrasound technician.