All researchProject overview

Medical imaging · 2022–2023

Grading prenatal hydronephrosis from ultrasound

Combining multiple fetal kidney ultrasound views into one severity grade, with the kidney outlined before classification.

Hamad Medical Corporation, Sidra Medicine and Qatar University

Five fetal kidney ultrasound images, one for each severity grade from SFU 0 to SFU IV, each shown above its annotated kidney region. The dark, fluid-filled area inside the kidney is larger at higher grades.
Figure 3 from Mahmud et al. (2024). Ultrasound views from five antenatal subjects, one for each Society for Fetal Urology grade, shown above the annotated kidney regions used for grading.View full-size figureSource article© 2024 Elsevier Ltd · author reuse

Approach

Can a pipeline use segmented kidneys across several ultrasound views to estimate a subject-level severity grade?

The study collected ultrasound images from 164 subjects. After curation, 2,062 images from 163 subjects were annotated for the kidney region. The workflow locates the region, outlines it with KUSNet, and predicts one of five severity grades with PHCNet. Weighted voting combines grades across ultrasound views at the subject level.

  1. Locate the kidney region
  2. Segment with KUSNet and grade with PHCNet
  3. Combine views into one grade

Illustrated example

From ultrasound views to one severity grade

  1. Locate the kidney

    Find the kidney region in each view.

  2. Segment and grade each view

    KUSNet outlines the kidney. PHCNet grades the segmented views. In this example all three give grade II.

  3. Combine the grades

    Weighted voting gives one subject-level result: Society for Fetal Urology grade II.

Original schematic based on the study’s methods. The views and example grades are illustrative. Study methods

Reported results

KUSNet segmented the kidney with a 92.1% Dice score. With segmented views, five-grade classification reached 93.9% subject-level accuracy and an 89.0% F1 score. The two tasks were evaluated separately.

Kidney segmentation Dice
92.1%
Subject-level grading accuracy
93.9%
Subjects in the curated cohort
163
Grid of twelve kidney ultrasound views with the ground-truth kidney mask and the masks predicted by UNet++, UNet3+ and KUSNet beneath each view.
Figure 6 from Mahmud et al. (2024). Kidney masks predicted by UNet++, UNet3+ and KUSNet, compared with the ground truth for twelve ultrasound views (columns a–l).View full-size figureSource article© 2024 Elsevier Ltd · author reuse

My contribution

I contributed to study design and data curation, developed the methods and software, and wrote the original draft with the research team.

Study scope

Segmentation and grading used five subject-independent folds on the retrospective QU–HMC–AHN cohort. The earlier region-detection stage used a separate two-fold evaluation. Reported grading performance reflects the curated views and reference grades in that cohort.