An Explainable Deep Learning Framework for Automated Fetal Ultrasound Standard-Plane Classification: Integrating ResNet50, Grad-CAM, and End-to-End Clinical Deployment

Dheekshitha Jain M J; Supreetha Gowda HD

An Explainable Deep Learning Framework for Automated Fetal Ultrasound Standard-Plane Classification: Integrating ResNet50, Grad-CAM, and End-to-End Clinical Deployment

Author(s)	Ms. Dheekshitha Jain M J, Dr. Supreetha Gowda HD
Country	India
Abstract	Reliable interpretation of fetal ultrasound imagery continues to depend heavily on the availability of adequately trained sonographers, a constraint that is most severe in low-resource settings. In large parts of rural India and other low- and middle-income regions, ultrasound hardware is increasingly accessible, yet the shortage of qualified personnel to acquire and interpret standard diagnostic planes remains a major barrier to consistent prenatal care. This paper presents FetalAI, an end-to-end deep learning framework for automated classification of fetal ultrasound images into four diagnostically standard anatomical planes — brain, abdomen, femur, and thorax. The system is built around a ResNet50 convolutional neural network fine-tuned through a two-phase transfer-learning protocol on the publicly available Fetal Planes dataset (Zenodo Record 3904280), achieving a macro-averaged validation accuracy of 97.4% across the four target classes. To mitigate the interpretability limitations that restrict clinical trust in deep learning systems, Grad-CAM-based visual explanations are generated for every prediction; qualitative inspection confirms that the resulting attention maps consistently localize to anatomically recognized landmarks such as the falx cerebri, the femoral diaphysis, and the four-chamber cardiac silhouette. Beyond classification, FetalAI incorporates a three-criterion automated input-validation stage that filters non-ultrasound or low-confidence inputs prior to inference, an automated PDF diagnostic report generator, an auditable prediction log, and a Streamlit-based web interface intended for use by non-specialist healthcare personnel. Comparative evaluation against five recently published systems and two internal baselines indicates that FetalAI's classification accuracy is competitive with — though not the highest among — published research prototypes, but that it is the only reviewed system to combine explainability, automated reporting, input safeguarding, and a deployable interface within a single platform, characteristics that are arguably more decisive for real-world clinical adoption than benchmark accuracy alone.
Keywords	Fetal Ultrasound Image Classification; ResNet50; Transfer Learning; Grad-CAM; Explainable Artificial Intelligence; Medical Image Analysis; Prenatal Healthcare Informatics; Clinical Decision Support; Streamlit Deployment.
Field	Computer Applications
Published In	Volume 17, Issue 7, July 2026
Published On	2026-07-04

View / Download PDF File

About IJTAS Fees & Payment Current Issue Publication Archive	Submit Research Paper Track Submission Status Publication Guidelines Publication Ethics	Join as a Reviewer Editors & Reviewers Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us	Message on WhatsApp	+91-9687-182-185	editor@ijtas.com

International Journal of Technology and Applied Science

A Widely Indexed Open Access Peer Reviewed Multidisciplinary Bi-monthly Scholarly International Journal

An Explainable Deep Learning Framework for Automated Fetal Ultrasound Standard-Plane Classification: Integrating ResNet50, Grad-CAM, and End-to-End Clinical Deployment

Share this