Manuscript received July 25, 2025; revised October 13, 2025; accepted November 24, 2025; published May 27, 2026.

Abstract—Ultrasound imaging is essential for early pregnancy assessment; however, manual interpretation remains limited by operator dependency and speckle noise. Unlike previous studies that employed You Only Look Once (YOLO)v5, YOLOv6, or YOLOv7, this study introduces an automated approach for detecting key fetal head anatomical structures during the first trimester based on the YOLOv8 architecture. The study systematically compares raw ultrasound images with Hybrid Speckle Noise Reduction (HSNR) preprocessing to evaluate the trade-off between visual enhancement and fine-structure preservation, as quantified by a Structural Similarity Index Measure (SSIM) of 0.912 and a Peak Signal-to-Noise Ratio (PSNR) of 29.8 decibels. Semi-automated annotation with expert validation, yielding a Cohen’s kappa value of 0.84, ensured high labeling reliability. YOLOv8 was trained under multiple configurations, including different optimizers such as stochastic gradient descent, Adam, and AdamW, combined with early stopping and stratified five-fold cross-validation at a resolution of 320 by 320 pixels to balance anatomical detail and real-time efficiency. The optimal configuration using AdamW with early stopping achieved strong detection performance, with a mean average precision at an Intersection-over-Union (IoU) threshold of 0.50 of 0.907 and a recall of 0.857, outperforming models trained on images processed with speckle noise reduction. While speckle noise reduction improved overall image clarity, it slightly reduced the detectability of subtle anatomical features such as nuchal translucency and nasal skin due to excessive smoothing. The proposed model achieved an inference speed of 45 frames per second on a graphics processing unit, demonstrating its feasibility for real-time clinical deployment. Overall, these results highlight the potential of an optimized YOLOv8 model trained on raw ultrasound images as an efficient, reliable, and clinically applicable approach for artificial intelligence–assisted prenatal screening.

Keywords—ultrasound image analysis, fetal head detection, first-trimester screening, deep learning, prenatal imaging, You Only Look Once (YOLO)v8, fetal head anatomical structures

Cite: Fajar A. Hermawati and Danara D. Caesa, "Automatic Detection of Key Fetal Head Anatomical Structures in the First Trimester Using Ultrasound Images," Journal of Image and Graphics, Vol. 14, No. 3, pp. 412-424, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).