Publication

Abstract : Ultrasound (US) imaging is often employed for monitoring fetal development throughout pregnancy. However, the manual detection of fetal anatomy presents several challenges to clinicians and healthcare professionals, including the structural similarity of fetal anatomical features, the position of the fetus, and the expertise of the sonographer. Artificial intelligence (AI) is now playing a significant role in developing AI-assisted tools in medical imaging to help healthcare providers and can aid in addressing challenges associated with fetal anatomy detection. Therefore, this article proposes a spatial attention (SA) deployed convolutional neural network (CNN) called VGGSA for efficient multiclass classification of the generally used fetal biometry planes during routine examinations. A pretrained VGG-19 CNN model is utilized as a deep feature extractor in VGGSA. The proposed VGGSA network integrates an SA module before the final pooling layer to enhance the feature representation capability of the backbone feature extractor. Leveraging the attention module in CNNs helps reduce misinterpretations caused by the inherent anatomical structural similarity between standard and nonstandard fetal organs. The attention module enables the model to focus on significant regions of the images, resulting in improved classification performance. The experiments utilized two publicly available fetal US datasets to evaluate the efficacy of the proposed VGGSA network. Experimental results demonstrate that the proposed work surpasses the state-of-the-art deep learning (DL) models. The Grad-CAM technique is also applied to visualize the predictive nature of the VGGSA network.