A Dual Adversarial Calibration Framework for Automatic Fetal Brain Biometry

TL;DR

This paper introduces a dual adversarial calibration framework that enables accurate fetal brain biometry on low-cost ultrasound images by adapting models trained on high-end images, significantly reducing error rates.

Contribution

The novel unsupervised domain adaptation method employs dual adversarial pathways to make deep models invariant to image distribution shifts between high-end and low-cost ultrasound images.

Findings

Achieved mean absolute errors of 2.43mm for transcerebellar diameter and 1.65mm for head circumference.

Outperformed state-of-the-art methods with errors of 7.28mm and 5.65mm respectively.

Demonstrated effectiveness in low-resource settings for fetal biometry.

Abstract

This paper presents a novel approach to automatic fetal brain biometry motivated by needs in low- and medium- income countries. Specifically, we leverage high-end (HE) ultrasound images to build a biometry solution for low-cost (LC) point-of-care ultrasound images. We propose a novel unsupervised domain adaptation approach to train deep models to be invariant to significant image distribution shift between the image types. Our proposed method, which employs a Dual Adversarial Calibration (DAC) framework, consists of adversarial pathways which enforce model invariance to; i) adversarial perturbations in the feature space derived from LC images, and ii) appearance domain discrepancy. Our Dual Adversarial Calibration method estimates transcerebellar diameter and head circumference on images from low-cost ultrasound devices with a mean absolute error (MAE) of 2.43mm and 1.65mm, compared with 7.28 mm and 5.65 mm respectively for SOTA.