# Validity of intracerebral haemorrhage volume assessment: comparison of fully automated segmentation analysis with manual ABC/2 and semi-automated measurement

## TL;DR

This study compares automated, semi-automated, and manual methods for measuring brain hemorrhage volume and finds significant disagreement between them.

## Contribution

The study evaluates the validity of automated ICH volume measurement against semi-automated and manual methods.

## Key findings

- StrokeViewer showed a mean absolute difference of 11.2 mL compared to the reference standard.
- ABC/2 had a mean absolute difference of 10.1 mL compared to the reference standard.
- Neither automated nor manual methods consistently agreed with the reference within 10%.

## Abstract

In patients with spontaneous Intracerebral haemorrhage (ICH), ICH volume is associated with neurological and functional outcome and it is an important factor in neurosurgical decision-making. This study aims to assess the agreement between automated ICH volume measurement, the ABC/2 method and semi-automatic segmentation.

We retrospectively collected data from 300 consecutive adult patients (November 2018–June 2023) with spontaneous, supratentorial ICH, with a symptom onset-to-CT time ≤ 24 hours. We measured ICH volumes with automated StrokeViewer software, the manual ABC/2 method and semi-automated Brainlab software (reference standard). We performed a Bland–Altman analysis to compare measurements, considering a deviation of 10% or less from the reference standard as clinically acceptable.

The median age was 69 years (IQR, 57–76), 124 (41.3%) were women and the median NIHSS was 18 (IQR, 11–23). Median ICH volume was 26.0 mL (IQR, 9.3–59.2) (Brainlab). StrokeViewer also segmented hyperdense structures other than ICH, but occasionally, it only segmented part of the ICH accurately. The mean absolute differences were 11.2 mL (limits of agreement [LoA] −34.3 to 56.7) between StrokeViewer and Brainlab, −0.42 mL (LoA −65.0 to 64.9) between StrokeViewer and ABC/2 and 10.1 mL (LoA −36.2 to 56.4) between ABC/2 and Brainlab.

There is substantial disagreement between the 3 methods for the measurement of ICH volume. Considering a clinical limit of acceptance of 10% or less, neither StrokeViewer nor ABC/2 agreed with our reference standard. Therefore, StrokeViewer results should not be used for volume-based clinical decisions without visual confirmation of adequate segmentation.

Graphical abstract