Unpaired MR-CT brain dataset for unsupervised image translation

Omar S Al-Kadi; Israa Almallahi; Alaa Abu-Srhan; A M Mohammad Abushariah; Waleed Mahafza

doi:10.1016/j.dib.2022.108109

Unpaired MR-CT brain dataset for unsupervised image translation

Data Brief. 2022 Mar 30:42:108109. doi: 10.1016/j.dib.2022.108109. eCollection 2022 Jun.

Authors

Omar S Al-Kadi¹, Israa Almallahi², Alaa Abu-Srhan³, A M Mohammad Abushariah¹, Waleed Mahafza²

Affiliations

¹ King Abdullah II School of Information Technology, The University of Jordan, Amman 11942, Jordan.
² Department of Diagnostic Radiology, Jordan University Hospital, Amman 11942, Jordan.
³ Department of Basic Science, The Hashemite University, Zarqa, Jordan.

Abstract

The data presented in this article deals with the problem of brain tumor image translation across different modalities. The provided dataset represents unpaired brain magnetic resonance (MR) and computed tomography (CT) image data volumes of 20 patients. This includes 179 two-dimensional (2D) axial MR and CT images. The MR cases are acquired using Siemens Verio scanner, while the CT images with a Siemens Somatom scanner. The MR and CT tumor volumes were collected, diagnosed and annotated by experienced radiologists specialized in oncology and radiotherapy. The collected image volumes can be useful for researchers working in the field of artificial intelligence (AI) applications for brain tumor detection, classification and segmentation in MR and CT modalities. The provided tumor masks per each tumor volume can assist data scientists with limited background in cancer imaging. Moreover, clinical interpretation is given per each tumor volume, which can assist in deep learning model training with multiple source data (non-imaging or textual data) as well. The provided dataset can facilitate for annotation-efficient lesion segmentation using bidirectional MR-CT cross-modality image translation.

Keywords: Brain tumors; Image translation; MR and CT image dataset; Unsupervised learning.