fidopt Interactive view
CT to PET, synthesis

ENHANCE.PET 1.6k, CT to PET

The CT (source) and the FDG-PET in SUV (target) of one whole-body PET/CT session of a patient with lung cancer at University Hospital Leipzig, on the PET's grid (ENHANCE.PET 1.6k).

30 test caseswhole bodysource CT, target PET
Look at the images
Read before comparing (7)
  • Pool. University Hospital Leipzig's 384 lung cancer FDG-PET/CT studies of the release (CC BY 4.0), one study a patient by the rule of ../EnhancePETBodyCT/prepare.py (three exams delivered twice are taken once). 180 of Careggi's 199 PETs are in scanner counts (PT-details.xlsx's Units, CNTS; the other 19 in Bq/mL) with no factor in the files to bring them to Bq/mL, and the autoPET studies are already AutoPET_CTPET_Syn's: both are left out.
  • Alignment. CT and PET of one session on one PET/CT scanner in one frame of reference (the provider's NIfTI); the CT is brought onto the PET's grid through that frame, as AutoPET_CTPET_Syn does. On the three test cases looked at, bladder, kidneys, liver and heart fall together; breathing moves the diaphragm between the two, as in any PET/CT.
  • Target. The PET's Bq/mL in body-weight SUV, the dose decayed from injection to the series time given in PT-details.xlsx, assuming the scanner corrects every slice to the start of the series, as Siemens scanners do; the spreadsheet does not say. Uptake times run from 46 to 152 minutes (median 88) on the test cases, so the same organ reads differently from case to case.
  • Images. PET 168 x 168 voxels of 4.06 mm, slices 2 or 3 mm; CT resampled onto it, -1024 outside the CT. The faces are blanked by the provider in both: the PET reads 0 in a box over the head, which the mask leaves out with the PET's other zeros.
  • Overlap. These are Leipzig patients other than EnhancePETBodyCT's; no other dataset of the corpus holds Leipzig studies.
  • Start. On the test cases, at a range of 20 SUV, a constant at the target's median in the body scores a median MAE of 0.47 SUV (0.38 to 0.71), a PSNR of 27.4 dB (15.9 to 29.8) and an SSIM of 0.71; the CT as it is, on another scale, scores nothing. The body's median SUV is 0.63; bladder, brain, heart and tumours carry the signal.
  • Outside numbers. Romoli 2026 reports CT to PET (SSIM 0.95) on its own split of the 583 Leipzig and Careggi lung studies, which may include these test patients; not comparable.

No method has a result on every case of this dataset yet.

Case by case

Click a score to look at that result: the target image, and over it the method's synthetic image. Underlined scores have their image; for the others you see the case without the result.

About the data

Test cases
30 test patients drawn with seed 2026 among Leipzig's 299 patients outside development and EnhancePETBodyCT, then 30 development patients, the last of the same seeded order. One study a patient, by the rule of ../EnhancePETBodyCT/prepare.py; Careggi is left out as a whole: 180 of its 199 PETs are in scanner counts the files give no factor for. The release has no test set.
Truth
target image as released the session's FDG-PET in body-weight SUV, from the provider's NIfTI in Bq/mL with the weight, dose and times of PT-details.xlsx, on its own grid; the CT is brought onto it through the scanner's frame
Licence
CC BY 4.0 (Leipzig studies)
30 cases for tuning, never tested

1385, 1033, 1139, 1225, 1245, 1138, 1249, 1085, 1360, 1110, 1160, 1148, 1096, 1101, 1338, 1199, 1370, 1270, 1252, 1190, 1140, 1130, 1049, 1077, 1180, 1323, 1320, 1102, 1133, 1144