# Fractal-geometry analysis of pediatric posterior fossa tumors – a preoperative tool for prediction of histopathology

## TL;DR

This study uses fractal geometry to analyze MRI images of pediatric brain tumors, helping predict their type before surgery.

## Contribution

The study introduces fractal geometry as a novel radiological biomarker for preoperative histopathological prediction in pediatric posterior fossa tumors.

## Key findings

- Fractal metrics showed significant differences among pilocytic astrocytomas, embryonal tumors, and ependymomas.
- A prediction formula using fractal dimension and FLAIR lacunarity achieved an AUC of 0.793 for tumor classification.
- Tumor volume and cystic component were significant factors in differentiating tumor types.

## Abstract

The most common types of pediatric posterior fossa tumors are pilocytic astrocytomas, embryonal tumors, and ependymomas. Preoperative knowledge of the pathological diagnosis may facilitate optimal surgical management. The complex geometry of tumors can be characterized by using different metrics, such as fractal geometry parameters that play an important role in the description of irregular, rough shapes. Our aim was to perform a fractal analysis of pediatric posterior fossa tumors and identify potential new radiological biomarkers. We conducted a retrospective clinical study using preoperative MRI images and clinical data from pediatric patients who had undergone surgery for posterior fossa tumors. T1, ceT1, T2 and FLAIR sequences were acquired for all patients. The tumors were segmented using ITK-SNAP software, then fractal analysis, t-tests, Fischer exact tests, logistic regression, and ROC analysis were performed. Forty-four patients met the selection criteria. T-tests evaluating tumor volume, fractal dimension (FD), FLAIR lacunarity index (LI), and the Fischer test assessing the cystic component revealed significant differences among the three main tumor types. The prediction formula was constructed by applying logistic regression to weight the selected factors, achieving an AUC of 0.793 (95% CI = 0.628–0.921). Our study suggests that fractal metrics are useful tools for the preoperative estimation of histological diagnosis.

The online version contains supplementary material available at 10.1007/s10143-025-04078-9.