Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2026
|
| Online Access: | https://doi.org/10.5281/zenodo.19702395 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- <p>Dataset of cervical MRI protocol Sagital T1 = C-ST1</p> <p>Dataset of cervical MRI protocol Sagital T2 = C-ST2</p> <p>Dataset of lumbar spine MRI protocol Sagital T1 = LS-ST1</p> <p>Dataset of lumbar spine MRI protocol Sagital T2 = LS-ST2</p> <p> </p> <p>This dataset was constructed in collaboration with the Radiology Service and the Medical Physics Service at Hospital de Clínicas de Porto Alegre (HCPA), located in South Brazil. The study was reviewed and approved by the Research Ethics Committee of HCPA (CAAE number 74933423.2.0000.5327).</p> <p>Using the Enterprise Viewer 8.12, a database query was conducted to retrieve MRI examinations based on the following criteria: Philips Achieva 1.5T equipment, cervical and lumbosacral spine imaging, patients aged 18 years or older, and examinations performed between January 1, 2016, and October 31, 2023. Incomplete examinations were excluded, and all data were anonymized using DicomCleaner™. The final dataset comprised 668 lumbosacral and 679 cervical spine MRI examinations. </p> <p>A Python script was developed to automate slice selection, categorizing images based on examination type, acquisition protocol, coil, and acquisition plane. The largest subsets identified were sagittal T1 (ST1) and T2 (ST2) protocols for both cervical and lumbosacral spine regions, yielding datasets comprising 292 samples for lumbosacral spine (LS) ST1, 237 for LS ST2, 374 for cervical spine (C) ST1, and 357 for C ST2.</p> <p>Image quality metrics were calculated using entropy power and spectral flatness. Entropy power quantifies the uncertainty or variability of a signal's distribution. In the context of images, higher values may indicate greater randomness or complexity, which could correspond to noisier patterns. Spectral flatness is a frequency-domain measure that assesses the uniformity of a signal's power spectrum. Higher values indicate that the signal's spectrum resembles white noise, potentially corresponding to less structured regions in an image. </p> <p>Finally, the target variable was defined by normalizing entropy power and spectral flatness values. Image quality was categorized relative to the median, with lower target values corresponding to higher-quality images (class 1) and higher values indicating lower-quality images (class 0). At the end of the data collection step, four datasets -- LS-ST1, LS-ST2, C-ST1, and C-ST2 -- were prepared for subsequent analyses.<br> </p>