# Noninvasive prediction of Ki-67 expression level in IDH-wildtype glioblastoma using MRI histogram analysis: comparison and combination of MRI morphological features

## TL;DR

This study shows that MRI scans can noninvasively predict Ki-67 levels in a type of brain cancer called IDH-wildtype glioblastoma, with better accuracy when combining different MRI features.

## Contribution

The novel contribution is demonstrating that combining MRI morphological features and histogram analysis improves noninvasive prediction of Ki-67 expression levels in IDH-wildtype glioblastoma.

## Key findings

- Ki-67 high-level expression group showed higher histogram parameters like min, P01, P50, and P75 compared to the low-level group.
- Combining MRI morphological features and histogram parameters achieved an AUC of 0.867 for predicting Ki-67 levels.
- Strong positive correlations were found between histogram parameters and Ki-67 expression levels.

## Abstract

To assess and compare the effectiveness of magnetic resonance imaging (MRI) morphological features and MRI histogram analysis in noninvasively predicting Ki-67 expression levels in patients with IDH-wildtype glioblastoma.

Forty-six cases of IDH-wildtype glioblastoma with measured Ki-67 expression levels from January 2022 to July 2024 were retrospectively collected. They were divided into Ki-67 low-level expression group (Ki-67<20%, n=20) and Ki-67 high-level expression group (Ki-67≥20%, n=26) according to Ki-67 expression level. MRI morphological features were assessed and recorded. MRI histogram analysis were performed on contrast-enhanced T1-weighted images. Differences between these parameters were compared between the two groups. The diagnostic performance was assessed by the area under the receiver operating characteristic curve (AUC). Spearman correlation was used to evaluate the relationship between histogram parameters and Ki-67 expression level.

Hemorrhage was more prone to occur in the Ki-67 high-level expression group (P=0.017). The min, P01, P50, and P75 of IDH-wildtype glioblastoma Ki-67 high-level expression group were higher than those of the Ki-67 low-level expression group (P<0.00357). There was a significant positive correlation between the min (r=0.774), P01 (r=0.729), P50 (r=0.625), P75 (r=0.591), and Ki-67 expression level (P<0.05). The optimal diagnostic performance was obtained by combining MRI morphological features and histogram parameters, with an AUC of 0.867.

Both MRI morphological features and histogram parameters could predict the Ki-67 expression level in IDH-wildtype glioblastoma, and the combined model integrating MRI morphological features and histogram parameters can be an excellent imaging biomarker for noninvasively predicting Ki-67 expression levels in patients with IDH-wildtype glioblastoma.