Cost-Efficient Light-Weight YOLO V5_s for Whole Fetus Detection in Early-Stage Ultrasound Scans

• DOI: 10.31357/ait.v4i02.8026
• Publication Date: 2025-03-07T02:31:26Z
• Authors: U.B. Balagalla; O.V. Jayawardane; J.V.D. Jayasooriya; C. De Alwis; A. Subasinghe

Routine medical Ultrasound (US) scans are recommended for expectant mothers to monitor the health and growth of fetus. However, in rural areas developing underdeveloped countries face difficulties receiving timely due a lack expertise facilities. Consequently, maternal fetal deaths occur at higher rates these countries, especially first trimester. Novel concepts such as Virtual Doctors, Hospital Home, self- Internet Medical Things (IoMT) may address aforementioned problem effectively. Therefore, computational-efficient algorithms which support low-end smart devices should be introduced assist provide comfortable scans. In light above, this paper discusses computationally efficient YOLO V5_s detection first-trimester ultrasound images using highly diverse dataset including abnormal multiple pregnancies. The implemented model was compared with five benchmark models, namely, ResNet-50 MobileNet-based faster R-CNN, YOLO-n, YOLO-m YOLO-l. comparatively better than R-CNN. Even though YOLO-n is most inexpensive model, its mAP 0.709, low, hence cannot applied clinical set-up. YOLO-l has best performance F-1 score 0.978 0.751, respectively. YOLO-s also achieved 0.979 0.734. subjective test conducted verify set-up experts field more years experience. analysis test, assessed through Fleiss Kappa, suggests substantial agreement beyond chance (κ = 0.69), while Intraclass Correlation Coefficient (ICC) indicates modest reliability (ICC 0.7). findings endorse application real-time whole fetuses trimester reduced computational complexity further validation.