fidopt Interactive view
CT to PET, synthesis

Lung-PET-CT-Dx, CT to FDG-PET

CT (source) and FDG-PET in body-weight SUV (target) of one lung cancer patient's PET/CT session at the Second Affiliated Hospital of Harbin Medical University, the CT brought onto the PET grid through their shared scanner frame (Lung-PET-CT-Dx collection).

10 test casesthoraxsource CT, target PET
Look at the images
Read before comparing (9)
  • Pairs. One Siemens Biograph 64 mCT session a patient: the attenuation-corrected PET and the whole-body attenuation-correction CT ('CT WB 1.0 B30f', or '3.0' on five of the pool; no contrast agent in its header), in one frame of reference. prepare.py checks the shared FrameOfReferenceUID and pins both series. Nothing is registered: the CT is resampled onto the PET grid with a cubic B-spline, air outside it, and rounded to whole Hounsfield units.
  • Field. The collection keeps only part of most CTs, the chest: 160 to 290 mm of the PET's 720 to 980 on 17 of the 20 cases, where Lung_Dx-A0164 keeps all 849 mm, Lung_Dx-A0220 675 and Lung_Dx-G0037 399. Where the PET runs past the CT the source reads -1024 and the body is left out of the mask.
  • Target. Body-weight SUV from the PET's rescale values and its DICOM dose, weight and decay information (AutoPET_CTPET_Syn.suv_factor), 49 to 127 minutes after injection in the pool. The mask is the body in the shared field, lungs included (AutoPET_CTPET_Syn.body).
  • Motion. Patients breathed freely during both scans, and the PET lasts minutes where the CT lasts seconds: the lung bases and the heart disagree by an amount nobody measured.
  • Uptake. Every patient has a biopsied lung cancer (adenocarcinoma, small-cell or squamous, the letter in the patient's name); what the CT does not show counts against every method.
  • Range. A dynamic range of 20 SUV puts PSNR and SSIM on the scale of the other FDG-PET synthesis datasets.
  • Baseline. On the 10 test cases a constant at the target's median inside the mask scores, as the mean over the cases and its range, an MAE of 0.37 (0.31 to 0.44) SUV, a PSNR of 29.0 (25.2 to 33.0) dB and an SSIM of 0.78 (0.73 to 0.81).
  • Unused. The 85 other PET studies have no whole-body CT: 55 only a chest CT ('Thorax' B31f or B70f), 27 only the fused 'Range' reformats, 3 neither. The radiologists' tumour boxes and BAMF's automatic lung masks are not used.
  • Overlap. SAROS labels CTs of 17 Lung-PET-CT-Dx patients. SAROSBodyCT holds six of them: Lung_Dx-A0165 is held out here, the five others have no whole-body CT. SAROS also labels the CT of the test case Lung_Dx-A0220 (its case_737, a training fold), which SAROSBodyCT does not hold. No other descriptor names the collection.

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
10 test patients drawn with seed 2026, then 10 development patients, one session each. The pool is the 48 patients whose PET/CT study holds a whole-body CT and an attenuation-corrected PET; Lung_Dx-A0165, whose CT SAROSBodyCT shows, is held out.
Truth
target image as released the attenuation-corrected FDG-PET of the same PET/CT session in body-weight SUV, aligned by the scanner frame
Licence
CC BY 4.0
10 cases for tuning, never tested

Lung_Dx-B0034, Lung_Dx-G0041, Lung_Dx-G0035, Lung_Dx-G0053, Lung_Dx-A0178, Lung_Dx-G0033, Lung_Dx-A0202, Lung_Dx-G0050, Lung_Dx-A0222, Lung_Dx-A0251