Anonymized Imaging Data for Two Small Cohorts of Head and Neck Cancer Patients

Images for two separate
cohorts of fifteen head and neck cancer patients diagnosed with oral or oropharyngeal squamous cell carcinoma. Images for each patient were acquired
before the start of radiotherapy.

The first cohort, labeled “heterogeneous” (HET),
consisted of patients with images acquired at different institutions and is termed
“heterogeneous” because of the variety of acquisition scanners and parameters
used in image generation. HET cohort images include T2-weighted MRI only.

The second cohort, labeled “homogeneous” (HOM),
consisted of patients from a single prospective clinical trial with the same imaging
protocol (NCT04265430, PA16-0302) and is termed “homogeneous” because of the
consistency in both scanner and acquisition parameters used in image generation. HOM cohort images include T2-weighted MRI for all patients and Dixon T1-w Water Enhanced MRI and CT for a subset of 5 patients.

For each image, regions of interest (ROI)s of various healthy tissue types and anatomical
locations were manually contoured in the same relative area for five slices by
one observer (medical student) using Velocity AI v.3.0.1 (Atlanta, GA, USA), verified
by a physician expert (radiologist), and exported as DICOM-RT Structure Set
files. The ROIs were: 1. cerebrospinal fluid inferior (CSF_inf), 2. cerebrospinal
fluid middle (CSF_mid), 3. cerebrospinal fluid superior (CSF_sup), 4. cheek fat
left (Fat_L), 5. cheek fat right
(Fat_R), 6. nape fat inferior (NapeFat_inf), 7. nape fat middle (NapeFat_mid), 8.
nape fat superior (NapeFat_sup), 9. neck fat (NeckFat), 10. masseter left (Masseter_L),
11. masseter right (Masseter_R), 12. rectus capitus posterior major (RCPM), 13.
skull, and 14. cerebellum.

All images were retrospectively acquired
from the University of Texas MD Anderson Cancer Center clinical databases in
agreement with an Institutional Review Board approved protocol designed to collect
data from patients with multiple imaging acquisitions (RCR03-0800). The
protocol included a waiver of informed consent.

DICOM data was anonymized using an in-house Python script that implements the RSNA CRP DICOM Anonymizer software. All files have had any DICOM header info and metadata containing PHI removed or replaced with dummy entries.

Note: A zipped file was uploaded, once unzipped, there should be 2 separate folders corresponding to each cohort, which can then be used as inputs to the github code at: https://github.com/kwahid/MRI_Intensity_Standardization.