🔬 Datasets¶

📋 Single-staging whole-body FDG/PSMA PET/CT¶

ℹ️ Information¶

The FDG cohort comprises 1014 studies of 501 patients diagnosed with histologically proven malignant melanoma, lymphoma, or lung cancer, along with 513 negative control patients. The PSMA cohort includes pre- and/or post-therapeutic PET/CT images of male individuals with prostate carcinoma, encompassing images with (537) and without PSMA-avid tumor lesions (60). Notably, the training datasets exhibit distinct age distributions: the FDG UKT cohort spans 570 male patients (mean age: 60; std: 16) and 444 female patients (mean age: 58; std: 16), whereas the PSMA MUC cohort tends to be older, with 378 male patients (mean age: 71; std: 8). Additionally, there are variations in imaging conditions between the FDG UKT and PSMA MUC cohorts, particularly regarding the types and number of PET/CT scanners utilized for acquisition. The PSMA MUC dataset was acquired using three different scanner types (Siemens Biograph 64-4R TruePoint, Siemens Biograph mCT Flow 20, and GE Discovery 690), whereas the FDG UKT dataset was acquired using a single scanner (Siemens Biograph mCT).

📥 Download¶

We provide the merged data as NIfTI in nnUNet format which can be downloaded from fdat (120GB):

The download will contain the resampled FDG and PSMA data as NiFTI files. It also contains the files obtained by running the nnUNet fingerprint extractor and a splits file which we use to design/train our baselines.

🎥 PET/CT acquisition protocol¶

FDG dataset: Patients fasted at least 6 h prior to the injection of approximately 350 MBq 18F-FDG. Whole-body PET/CT images were acquired using a Biograph mCT PET/CT scanner (Siemens, Healthcare GmbH, Erlangen, Germany) and were initiated approximately 60 min after intravenous tracer administration. Diagnostic CT scans of the neck, thorax, abdomen, and pelvis (200 reference mAs; 120 kV) were acquired 90 sec after intravenous injection of a contrast agent (90-120 ml Ultravist 370, Bayer AG) or without contrast agent (in case of existing contraindications). PET Images were reconstructed iteratively (three iterations, 21 subsets) with Gaussian post-reconstruction smoothing (2 mm full width at half-maximum). Slice thickness on contrast-enhanced CT was 2 or 3 mm.

PSMA dataset: Examinations were acquired on different PET/CT scanners (Siemens Biograph 64-4R TruePoint, Siemens Biograph mCT Flow 20, and GE Discovery 690). The imaging protocol mainly consisted of a diagnostic CT scan from the skull base to the mid-thigh using the following scan parameters: reference tube current exposure time product of 143 mAs (mean); tube voltage of 100kV or 120 kV for most cases, slice thickness of 3 mm for Biograph 64 and Biograph mCT, and 2.5 mm for GE Discovery 690 (except for 3 cases with 5 mm). Intravenous contrast enhancement was used in most studies (571), except for patients with contraindications (26). The whole-body PSMA-PET scan was acquired on average around 74 minutes after intravenous injection of 246 MBq 18F-PSMA (mean, 369 studies) or 214 MBq 68Ga-PSMA (mean, 228 studies), respectively. The PET data was reconstructed with attenuation correction derived from corresponding CT data. For GE Discovery 690 the reconstruction process employed a VPFX algorithm with voxel size 2.73 mm × 2.73 mm × 3.27 mm, for Siemens Biograph mCT Flow 20 a PSF+TOF algorithm (2 iterations, 21 subsets) with voxel size 4.07 mm × 4.07 mm × 3.00 mm, and for Siemens Biograph 64-4R TruePoint a PSF algorithm (3 iterations, 21 subsets) with voxel size 4.07 mm × 4.07 mm × 5.00 mm.

⌛ Training cohort¶

Training cases: 1,014 FDG studies (900 patients) and 597 PSMA studies (378 patients)

FDG training data consists of 1,014 studies acquired at the University Hospital Tübingen and is made publicly available on TCIA in DICOM format:
and on fdat in NIfTI format:

PSMA training data consists of 597 studies acquired the LMU University Hospital Munich and will be made publicly available on TCIA in DICOM format. The combined PSMA and FDG data is available on fdat in NIfTI format:

If you use this data, please cite:

Gatidis S, Kuestner T. A whole-body FDG-PET/CT dataset with manually annotated tumor lesions 
(FDG-PET-CT-Lesions) [Dataset]. The Cancer Imaging Archive, 2022. DOI: 10.7937/gkr0-xv29

Jeblick, K., et al. A whole-body PSMA-PET/CT dataset with manually annotated tumor lesions
(PSMA-PET-CT-Lesions) (Version 1) [Dataset]. The Cancer Imaging Archive, 2024. DOI: 10.7937/r7ep-3x37

A case (training or test) consists of one 3D whole body FDG-PET volume, one corresponding 3D whole body CT volume, one 3D binary mask of manually segmented tumor lesions on FDG-PET of the size of the PET volume, and a simulated human click. CT and PET were acquired simultaneously on a single PET/CT scanner in one session; thus PET and CT are anatomically aligned up to minor shifts due to physiological motion. A pre-rocessing script for resampling the PET and CT to the same matrix size will be provided. In addition, the human interaction in the form of a foreground (lesion) and background click are pre-simulated (for training and test). The pre-simulated clicks for training are provided in Github together with a script for further (parametrized) click simulations.

🗃️ Data structure¶

|--- imagesTr
     |--- tracer_patient1_study1_0000.nii.gz  (CT image resampled to PET)

     |--- tracer_patient1_study1_0001.nii.gz  (PET image in SUV)
     |--- ...
|--- labelsTr
     |--- tracer_patient1_study1.nii.gz       (manual annotations of tumor lesions)

|--- dataset.json                             (nnUNet dataset description)
|--- dataset_fingerprint.json                 (nnUNet dataset fingerprint)

|--- splits_final.json                        (reference 5fold split)

|--- psma_metadata.csv                        (metadata csv for psma)

|--- fdg_metadata.csv                         (original metadata csv for fdg)

⚙️ Data pre-processing¶

Please note, that the submission and evaluation interfaces provided by grand-challenge are working with .mha data. Hence, you will need to read the test images in your submission from an .mha file. We already provide interfaces and code for this in the baseline algorithms.

✒ Annotation¶

FDG PET/CT training and test data from UKT was annotated by a Radiologist with 10 years of experience in Hybrid Imaging and experience in machine learning research. FDG PET/CT test data from LMU was annotated by a radiologist with 8 years of experience in hybrid imaging. PSMA PET/CT training and test data from LMU as well as PSMA PET/CT test data from UKT was annotated by a single reader and reviewed by a radiologist with 5 years of experience in hybrid imaging.

The following annotation protocol was defined:
Step 1: Identification of tracer-avid tumor lesions by visual assessment of PET and CT information together with the clinical examination reports.
Step 2: Manual free-hand segmentation of identified lesions in axial slices.

📋 DeepPSMA¶

ℹ️ Information¶

autoPET V aligns with the DeepPSMA challenge, which provides a multi-center whole-body PSMA PET/CT (a mix of 68Ga-PSMA-617 and 18F-DCFPyL) cohort in 100 patients acquired under clinically routine conditions for prostate cancer imaging. DeepPSMA data encompass scans from different PET/CT systems and PSMA tracers, with expert-annotated prostate cancer lesions and a wide range of disease burden and physiological uptake patterns. Imaging protocols and reconstruction follow standard clinical practice, enabling complementary coverage of PSMA-specific uptake characteristics and scanner variability.

📥 Download¶

DeepPSMA is available in NIfTI format:

🗃️ Data structure¶

For each case, a subdirectory for the two tracers with relevant image data and annotations are given (CT, PET in units of SUV, and Total Tumor Burden). The threshold used for contouring each PET series is given in the threshold.json file. Only voxels above this value should be considered for labelling of disease. For PSMA, all cases use the same threshold value of 3, while FDG uses a variable liver-based value typically ranging from 2.5-5. We provide the output of Total Segmentator on the CT as well as a rigid registration parameter file which may be used to coarsely align PSMA and FDG images.

|--- train_0001
     |---PSMA
         |--- CT.nii.gz
         |--- PET.nii.gz
         |--- TTB.nii.gz
         |--- threshold.json
         |--- totseg_24.nii.gz
         |--- rigid.tfm - (FDG to PSMA)
     |--- FDG
         |--- CT.nii.gz
         |--- PET.nii.gz
         |--- TTB.nii.gz
         |--- threshold.json
         |--- totseg_24.nii.gz
         |--- rigid.tfm - (PSMA to FDG)
|--- train_0002
|--- train_0003
...

🎥 PET/CT acquisition protocol¶

Image data are acquired on standard diagnostic PET/CT systems. Predominantly from institutional scanners GE Discovery 710 & 690 PET/CTs, Siemens Biograph PET/CT, and Siemens Vision 600 PET/CT. Whole body (predominantly vertex to thighs) PET images with low-dose CT component for attenuation correction and anatomical localization. PET images reconstructed with standard corrections EANM EARL-compliant resolution recovery settings.

⌛ Training cohort¶

Training cases: 100 studies (100 patients)

⚙️ Data pre-processing¶

Please refer to the DeepPSMA baseline algorithm for further information on data pre-processing and usage.

✒ Annotation¶

Images are contoured based on a fixed SUV threshold per scan. All PSMA PET/CTs utilise SUV>=3 while FDG PET/CT images use a liver-based threshold (generally Liver Mean + 2*SD). The designated threshold for each study is provided with the image data. Subsequently, determination of malignant versus physiological areas of tracer avidity has been manually annotated by expert nuclear medicine physician with 5 or more years specialisation. All TTB labels are matched to the resolution of the PET image which is generally of slightly lower resolution than the CT component.

The manual annotation workflow is further detailed in:

Buteau JP, Martin AJ, Emmett L, Iravani A, Sandhu S, Joshua AM, et al. 
PSMA and FDG-PET as predictive and prognostic biomarkers in patients given [177Lu] Lu-PSMA-617 versus cabazitaxel for metastatic castration-resistant prostate cancer (TheraP): a biomarker analysis from a randomised, open-label, phase 2 trial. 
The Lancet Oncology. 2022;23(11):1389-97

📋 Longitudinal CT screening¶

ℹ️ Information¶

In the context of this challenge, the longitudinal CT dataset is provided as optional auxiliary data. It is not part of the primary evaluation task and is not explicitly assessed. Participants may use this data at their discretion (e.g. for pretraining, feature learning, or data-centric approaches), but its use is neither required nor directly evaluated for the PET/CT segmentation task.
The cohort consists of melanoma patients undergoing longitudinal CT screening examinations in an oncologic context for diagnosis, staging, or therapy response assessment. The CT cohort comprises whole-body imaging in 300 patients (female: 170, mean age: 64y, std age: 15y) of two imaging timepoints: baseline staging, and follow-up scans after therapy treatment. Training data was acquired at a single site (UKT).

📥 Download¶

We provide the data as NIfTI format which can be downloaded from fdat (54GB):

🎥 CT acquisition protocol¶

Patients were scanned with the inhouse whole-body staging protocol for a scan field from skull base to the middle of the femur with patients laid in a supine position, arms raised above the head. Scanning was performed during the portal-venous phase after administration of body-weight adapted contrast medium through the cubital vein. Attenuation-based tube current modulation (CARE Dose, reference mAs 240) and tube voltage (120 kV) were applied. The following scan parameters were used:
SOMATOM Force: collimation 128 × 0.6 mm, rotation time 0.5 s, pitch 0.6
Sensation64: collimation 64 × 0.6 mm, rotation time 0.5 s, pitch 0.6
SOMATOM Definition Flash: collimation 128 × 0.6 mm, rotation time 0.5 s, pitch 1.0
SOMATOM Definition AS: collimation 64 × 0.6 mm, rotation time 0.5 s, pitch 0.6
Biograph128: collimation 128 × 0.6 mm, rotation time 0.5 s, pitch 0.8
Slice thickness as well as increment were set to 3 mm. A medium smooth kernel was used for image reconstruction.

⌛ Training cohort¶

Training cases: 300 studies (300 patients)

Annotated longitudinal CT of two imaging time points in 300 studes was acquired at the University Hospital Tübingen and is made publicly available on fdat in NIfTI format:

🗃️ Data structure¶

The CTs are provided in Hounsfield units. The mask is in integer indicating the individual lesions. Filenames start with a unique patient ID (10 digits). One patient can be associated to multiple baseline and follow-up CT images. The training and test database have the following structure:

|--- inputsTr
  |--- c6f057b865.csv                          (lesion information for patient)     
  |--- c6f057b865_BL_00.json                   (lesion center of gravity per lesion in baseline CT; Grand-Challenge JSON format)            
  |--- c6f057b865_BL_img_BL_img_00.nii.gz      (CT baseline image)     
  |--- c6f057b865_BL_mask_BL_img_00.nii.gz     (CT baseline lesion mask, integer mask)        
  |--- c6f057b865_FU_00.json                   (lesion center of gravity per lesion in first follow-up CT; Grand-Challenge JSON format)     
  |--- c6f057b865_FU_01.json                   (lesion center of gravity per lesion in second follow-up CT; Grand-Challenge JSON format; if available)     
  |--- c6f057b865_FU_img_FU_img_00.nii.gz      (CT follow-up image, first body region)     
  |--- c6f057b865_FU_img_FU_img_01.nii.gz      (CT follow-up image, second body region; if available)           
  |--- ...  
|--- targetsTr            
  |--- c6f057b865_FU_mask_FU_img_00.nii.gz     (CT follow-up lesion mask of first body region, integer mask)     
  |--- c6f057b865_FU_mask_FU_img_01.nii.gz     (CT follow-up lesion mask of second body region, integer mask; if available)       
  |--- ...

CSV file¶

The CSV file contains the following columns:

• lesion_id: Continous ID count in the respective patient
• cog_bl: Lesion center of gravity in baseline image as 3D pixel coordinates
• img_id_bl: baseline image ID (either 0 or 1)
• cog_propagated: Lesion center of gravity (as 3D pixel coordinates) propagated from baseline to follow-up scan using a conventional registration (not available for all lesions)
• cog_fu: Lesion center of gravitiy in follow-up image as 3D pixel coordinates
• img_id_fu: follow-up image ID (either 0 or 1)
• lesion_type: Anatomical lesion location

⚙️ Data pre-processing¶

Please note, that the submission and evaluation interfaces provided by grand-challenge are working with .mha data. Hence, you will need to read the test images in your submission from an .mha file. We already provide interfaces and code for this in the baseline algorithms.

✒ Annotation¶

All data were manually annotated by two experienced radiologists. To this end, tumor lesions were manually segmented on the CT image data using dedicated software.

The following annotation protocol was defined:
Step 1: Identification of tumor lesions by visual assessment of CT information together with the clinical examination reports.
Step 2: Manual free-hand segmentation of identified lesions in axial slices.
Step 3: Baseline and follow-up segmentations are viewed side-by-side to mark the matching lesions.

🚧 Preliminary test set¶

For the self-evaluation of participating pipelines, we provide access to a preliminary test set. The preliminary test set does not reflect the final test set. Algorithm optimization on the preliminary test set will not yield satisfactory results on the final test set!

The access to this preliminary set is restricted and only possible through the docker containers submitted to the challenge, and only available for a limited time during the competition. The purpose of this is that participants can check the implementation and sanity of their approaches.

📊 Final test set¶

The final test set consists of 200 whole-body PET/CT studies collected from four international centers (University Hospital Tuebingen, Ludwig-Maximilians-University Munich, Peter MacCallum Cancer Centre, University Hospital Essen). The cohort is designed to evaluate generalization, robustness, and interaction behavior under clinically realistic conditions.

• Total: 200 studies
• 50 cases per center
• Cases include a broad spectrum of disease presentations, including:
  • lesion-present cases with varying tumor burden
  • lesion-absent or low-uptake studies
  • clinically challenging cases with ambiguous or physiological uptake

To ensure fair evaluation, detailed information on tracer distribution, acquisition parameters, and case composition will not be disclosed prior to the challenge deadline.

Category 1: Simulated Interaction¶

All submitted algorithms will be evaluated in the 200 test cases with the evaluation metrics.

Category 2: Clinician-Driven Interaction¶

A dedicated subset of the test cohort is used for evaluation under the clinician-driven interaction regime:

• 20 cases in total (5 per center)
• Each case is annotated by two independent physicians at each center
• Interactive segmentation is performed beforehand on interaction outputs derived from the baseline model with particular focus on clinically challenging scenarios

All submitted algorithms will be evaluated in the 20 test cases with the evaluation metrics.