fidopt Interactive view
Lesions and tumours, segmentation

NSCLC-Radiomics, Lung1

The pretreatment CT of a patient with non-small cell lung cancer treated by radiotherapy at MAASTRO, with the primary tumour's gross volume outlined by a radiation oncologist (NSCLC-Radiomics, Lung1, TCIA version 4).

30 test caseslungimage CTresearch only
Look at the images
Read before comparing (6)
  • Pool. Lung1 is the 422 NSCLC patients of Aerts et al. (2014) treated at MAASTRO in Maastricht, stage I to IIIb, each with one pretreatment CT and its planning contours. Version 4 of the collection aligned the SEG and RTSTRUCT study UIDs with the CT and added a missing image to LUNG1-246, which repeats slice 39 at the same position with the same pixels: prepare.py reads that file once.
  • Truth. One radiation oncologist's clinical contour of the primary tumour, slice by slice; no agreement between readers is given for these patients (NSCLC-Radiomics-Interobserver1 measures it on 22 patients of the same study). Graded expert, a single reader's. Involved lymph nodes are not in GTV-1: a method that also finds them is charged for it. The lungs, heart, oesophagus and spinal cord in the SEG are not used: they are not drawn on every patient.
  • Cases. The test tumours hold 0.5 to 454 mL (median 49 mL); LUNG1-282's is a 0.5 mL nodule by the right hilum. All test CTs are 512 by 512 with 0.98 mm pixels and 3 mm slices, in Hounsfield units, -1024 at the lowest.
  • Overlap. PleThoraCT draws its 20 test and development patients from the same CTs; they are held out. NSCLC-Radiomics-Interobserver1 (22 patients, same study) exports its CTs through CERR with new IDs, 154 or 178 slices and dates of 2019, so nothing in the DICOM links them to Lung1: the 25 Lung1 patients whose sex, age within a year once rounded and clinical T, N and M stage match one of its patients (3 of them PleThoraCT's) are held out, so that a later Interobserver1 build shares no patient; seven of its patients match none. SAROSBodyCT leaves this collection out by licence; no other dataset of the corpus names it. LCTSC's institution 1 is MAASTRO too (4D CT exhale phases): none of its 16 CTs in the corpus is one of the 60 cases (voxel correlation 0.79 at most on the 22 pairs of one size and spacing), but no crosswalk tells whether a patient is in both.
  • Start. A mask of the whole image scores a Dice of 0.001 at the median of the test CTs, an Otsu mask 0.005 (0 to 0.055). TotalSegmentator 2.18.0's lung_nodules task writes no nodule on LUNG1-381 (a 58.7 mL tumour): it is not a tumour model.
  • Outside numbers. None: Lung1 has no challenge and no official split; its 422 CTs are a common training set, so a published network may have seen the test cases.

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

Case by case

Click a score to look at that result: the image, the truth in green and the method's map in red. Underlined scores have their image; for the others you see the case without the result.

About the data

Test cases
30 test CTs drawn with seed 2026 among the 346 patients outside development, and 30 development CTs. The pool is the 421 patients with a SEG (LUNG1-128 has none) less 45 held out: PleThoraCT's 20 patients, 22 others whose clinical data match a NSCLC-Radiomics-Interobserver1 patient, and the three CTs with missing slices. The collection has no test set.
Truth
contours by experts the segment GTV-1 of the patient's SEG, the primary tumour's gross volume a radiation oncologist outlined for treatment planning on the CT (the collection's RTSTRUCT, converted to DICOM SEG by dcmqi)
Licence
CC BY-NC 3.0
30 cases for tuning, never tested

LUNG1-407, LUNG1-338, LUNG1-107, LUNG1-386, LUNG1-259, LUNG1-371, LUNG1-383, LUNG1-049, LUNG1-003, LUNG1-050, LUNG1-403, LUNG1-195, LUNG1-211, LUNG1-146, LUNG1-096, LUNG1-117, LUNG1-035, LUNG1-067, LUNG1-300, LUNG1-174, LUNG1-126, LUNG1-235, LUNG1-352, LUNG1-281, LUNG1-289, LUNG1-269, LUNG1-239, LUNG1-141, LUNG1-028, LUNG1-399