gms | German Medical Science

Introduction: Due to their high data quality, clinical studies are attractive for secondary use. In this context, de-identification of imaging data is essential to uphold patient privacy [1] while at the same time providing researchers with necessary information about the data. In this work, we present the current methods for DICOM-conformant de-identification for clinical studies and secondary use along the example of Positron Emission Tomography-Computed Tomography (PET/CT) imaging. PET/CT is a nuclear medicine imaging procedure widely used in tumor diagnostics. It involves injecting the patient with a low-level radioactive tracer [2]. The Standardized Uptake Value (SUV) in PET/CT quantitatively assesses the metabolic activity of tissues, aiding in tumor evaluation. SUV is calculated by normalizing the tracer concentration to the injected dose of the tracer and the patient's weight, allowing for comparison across patients and scans. High SUV values indicate increased metabolic activity. Calculating SUV requires accurate values for Patient's Weight, Patient's Size, Patient's Sex, and specific time points [3], which are often de-identified [4].

Methods: We examined the DICOM standard with regard to de-identification. The standard defines thorough de-identification options, so called profiles [4]. DICOM images are composed of the pixel data and structured metadata. However, there are more than 5000 unique DICOM metadata items defined, including so-called private tags possibly containing arbitrary information [5] and thus potentially identifying data. Therefore, de-identification might pose a challenge, even without consideration of identifying information in pixel data, e.g. burnt-in identifiers. To evaluate PET/CT de-identification, we examined four publicly available datasets, focusing on the first DICOM image of the first patient. Four open access PET/CT datasets were assessed to determine how they had been de-identified.

Results: The open access datasets present with varying levels of compliance with the current DICOM de-identification standard. While most datasets specify their de-identification methods, one dataset does not, though it was confirmed to be de-identified. Another dataset follows an older standard without specific codes, and two others use a consistent de-identification approach that includes several specific measures. However, necessary data for SUV calculation is missing in the first dataset, whereas the fourth includes this information directly in the private tags.

Discussion: To ensure the availability of SUV, private tags related to the PET procedure can be retained. However, for most manufacturers the SUV must be calculated from patient characteristics, necessitating the presence of specific metadata. Given that this metadata contains more sensitive information than needed for evaluation, it might be needed to adapt the standard de-identification process for secondary use to minimize data exposure. If any direct or indirect identifying data elements are preserved, it is recommended to conduct a statistical analysis to assess the remaining risk of re-identification.

Conclusion: The DICOM standard defines comprehensive de-identification options, that cover the requirements for clinical trials with PET/CT. For secondary use, deviating from the standard may be necessary to prevent re-identification. Depending on the information provided by manufacturers in their private tags, certain metadata might not be required, but then might bear the risk of unnoticed identifying data.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.