Lumbar Spine Vertebral Compression Fractures (VCFs) Dataset: MRI T1-Weighted Images for Benign and Malignant Classification

This dataset was prepared for the study of classification of benign vertebral compression fractures (VCFs) secondary to osteoporosis and malignant VCFs secondary to neoplastic infiltration. The original study in which it was used aimed to assist in differentiating these conditions using three-dimensional radiomic techniques and artificial neural networks.

This dataset was assembled from sagittal T1-weighted magnetic resonance imaging (MRI) scans of the lumbar spine obtained from consecutive patients diagnosed with benign or malignant VCFs at the University Hospital of the Ribeirão Preto Medical School (HCFMRP) between the years 2010 and 2019. The images were acquired using the Philips Achieva 1.5 T and 3 T MRI systems and were stored in the DICOM (Digital Imaging and Communications in Medicine) format.

The compilation of the dataset followed a rigorous selection and filtering process. From the initial set of cases of vertebral fractures in the lumbar region, patients who had received prior treatment (such as chemotherapy, radiotherapy, or surgery), those with fractures of traumatic etiology, old fractures, patients under 18 years old, and cases of malignant fractures without biopsy confirmation were excluded. With these exclusions, the final set consists of 91 patients (36 men and 55 women, with a mean age of 64.24 ± 11.75 years), of which 47 have benign VCFs and 44 have malignant VCFs.

For the segmentation of fractured vertebrae, the images were pre-processed by normalizing the intensity to 256 gray levels (0 to 255), and histogram equalization was applied to improve contrast. The vertebrae were semi-automatically segmented using the 3D Slicer software. Each segmentation was saved in the "nrrd" format, native to 3D Slicer. The entire segmentation process, as well as the definition and application of exclusion criteria, were supervised by a senior radiologist with 20 years of experience in musculoskeletal radiology.

The structure of this dataset includes, in addition to the DICOM exams, a directory containing the requantized images to 256 gray levels in the "nrrd" format and another with the segmentation files in the "seg.nrrd" format. A spreadsheet with detailed information about each patient's class, sex, age, and which vertebral bodies were segmented is also available. All DICOM files in this dataset have been anonymized to ensure patient privacy.

This dataset was structured to provide a robust basis for the training and validation of machine learning models focused on the classification of vertebral compression fractures. It was designed to aid in the massive three-dimensional extraction of radiomic features, enabling the search for radiomic signatures capable of assisting radiologists in the accurate characterization of these fractures.

For more information on how the dataset was created and used, please refer to the original article: Chiari-Correia NS, Nogueira-Barbosa MH, Chiari-Correia RD, Azevedo-Marques PM. A 3D Radiomics-Based Artificial Neural Network Model for Benign Versus Malignant Vertebral Compression Fracture Classification in MRI. J Digit Imaging. 2023;36(4):1565-1577. doi:10.1007/s10278-023-00847-4