Three-Dimensional Postmortem Ultrasound of the Fetal Corneal Volume to Estimate Postmortem Interval

TL;DR

This study uses 3D ultrasound to measure fetal eye structures and accurately estimate the time since death using a machine learning model.

Contribution

A novel gradient boosting machine model using 3D ultrasound corneal and ocular volume ratios to estimate fetal postmortem interval.

Findings

The cornea-to-eyeball volume ratio is a strong independent predictor of postmortem interval (p < 0.001).

The GBM model achieved 91% variance explained in training and 75% in validation, with minimal overfitting.

A Shiny web application was developed to implement the model for clinical use.

Abstract

Objectives: To develop and prospectively validate a predictive model to estimate the fetal postmortem interval (PMI) using three-dimensional postmortem ultrasound (3D PM-US) measurements of corneal and ocular volumes. Methods: Single-center study including fetuses ≥ 20 weeks’ gestation with known time of death after feticide. A retrospective training cohort (n = 63; November 2022–July 2023) and a prospective validation cohort (n = 28; February–August 2025) were used. Corneal and ocular volumes were measured using the VOCAL™ rotation multiplanar technique; the cornea-to-eyeball volume ratio was calculated for each case. Automated machine learning (AutoML) was used to develop and validate a gradient boosting machine (GBM) model. Model performance was evaluated using the root mean squared error (RMSE), mean absolute error (MAE), and coefficient of determination (R2). Results: Ninety-four fetuses were included; three were excluded (two for extreme death–US intervals of 165 and 166 h; one for open eyelids). Median gestational age was 29.3 weeks (IQR 27.2–32.9); median birthweight was 1325 g (IQR 980–1880). The cornea-to-eyeball volume ratio was an independent predictor of PMI (p < 0.001). The GBM model explained 91% of the variance in the training cohort (R2 = 0.91, RMSE = 11.49 h, MAE = 8.45 h) and 75% in the validation cohort (R2 = 0.75, RMSE = 18.32 h, MAE = 14.49 h), demonstrating strong predictive accuracy and minimal overfitting. Variable importance analysis confirmed the cornea-to-eyeball ratio as the most influential and biologically plausible predictor of PMI. A Shiny web application was developed to facilitate clinical implementation. Conclusions: 3D PM-US measurements of the fetal cornea and eyeball can reliably and quantitatively estimate the PMI with good predictive accuracy using a GBM model. Multicenter studies are required to further refine the model, enable external validation, and determine its clinical and forensic utility.

Figures