Benchmark for Wilms' Tumor Segmentation

Wilms' tumor, or nephroblastoma is the most frequent malignant kidney tumor in childhood. About 75% of all patients are younger than five years - with a peak between two and three years. In Europe, diagnosis and therapy follow the guidelines of the International Society of Pediatric Oncology (SIOP). One of the most important characteristics of this therapy protocol is a preoperative chemotherapy. Clinicians categorize patients as high-, intermediate- or low-risk candidates according to histology, local stage and tumor volume after chemotherapy. Postoperative treatment ranges from no chemotherapy (low risk stage I) up to chemotherapy with irradiation of the tumor bed (high risk, stage II and III).
The most common histological subtypes of regressive and mixed type actually belong to the intermediate risk tumors. However, if, after chemotherapy, these tumors have a volume of more than 500ml, they are highly malignant and the patients are treated according to the high risk group protocol. In order not to expose children to unnecessary medical burden on the one hand and to maximize their chances of survival on the other, an exact determination of the tumor volume is indispensable.

The image data consists of 28 multi-sequence MR scans from 17 Wilms' tumor patients (5 male, 12 female), out of which 15 have been acquired from intermediate risk tumor (histological diagnosis: stromal predominant (2), mixed histology (6) or regressive type (7)) and 2 from high risk tumor types (histological diagnosis: blastemal predominant). For eleven patients, we have both, data before and after chemotherapy. The remaining ones are missing either data before or after chemotherapy, respectively.
Only patients with histologically confirmed Wilms' tumors were eligible for inclusion. The images were acquired at different centers over the course of several years, using MR scanners from different vendors, varying field strength (1.5T and 3T) and implementations of the imaging sequences. The data sets used in our benchmark share the following three MRI sequences:

• T2: T2-weighted images, axial 2D acquisition with 3.6mm to 9.1mm slice thickness and inslice-sampling ranging from 0.3mm to 1.4mm.
• T1: T1-weighted images, native image, axial 2D acquisition with 2.5mm to 9.1mm slice thickness and inslice-sampling ranging from 0.5mm to 1.6mm.
• T1c: T1c-weighted and contrast enhanced (Gadolinium) images, axial 2D acquisition with 1.8mm to 7.7mm slice thickness and inslice-sampling ranging from 0.5mm to 1.6mm.

The different MRI sequences were spatially co-registered on the T2 sequence using a rigid transformation.

Contact Person:	Sabine Müller
E-mail:	smueller -at- mia.uni-saarland.de
Postal Address:	Mathematical Image Analysis Group
	Faculty of Mathematics and Computer Science,
	Saarland University, Building E1.7
	66123 Saarbrücken, Germany
Office:	Room 4.04 Building E1.7, Saarbrücken Campus

MIA Group
©2001-2023
The author is not
responsible for
the content of
external pages.

Imprint - Data protection