fidopt Interactive view
Lumbar multifidus on ultrasound, 2D segmentation

LUMINOUS, lumbar multifidus in ultrasound

A transverse B-mode image of a young athlete's lower back at L5, 2D, with the lumbar multifidus outlined by hand on one or both sides (LUMINOUS, Concordia University).

10 test casesspineimage US2D, distances in mmresearch only
Look at the images
Read before comparing (8)
  • Images. 820 x 614 grey TIFFs from a GE LOGIQ e with a curvilinear probe at 5 MHz, gain 60 and 8 cm depth for every image, taken by one investigator; the position ('Prone Rest' or 'Standing Rest') is burnt into most images; subject_083's shows only its side. The pixel size, 10/57 mm, is the Fiji calibration every mask carries; the B-mode files carry none.
  • Truth. One rater outlined every muscle; no second reading of the masks is published, and the agreement measured is on the area, not the outline. Images where the muscle's landmarks could not be told were left out of the database.
  • Sides. A subject was imaged bilaterally, both muscles on one image, or one side an image when the muscles are large; one label holds the muscles of the case's image, so 23 of the 109 subjects' cases hold two muscles (2 of the 10 test cases) and 86 one.
  • Subjects. 109 Concordia varsity athletes, 64 men and 45 women aged 21.1 ± 1.7 years, each with 1 to 4 images (prone and standing, one or both sides); only the case's image is used.
  • Overlap. No other dataset of the corpus holds IMPACT Laboratory or Concordia images; SonoCorpus's MD5 manifest finds none of these images in its other 52 datasets.
  • Outside numbers. The masks are public since 2020. SonoBase was evaluated on LUMINOUS, not trained on it.
  • Start. On the 10 test images, the whole image scores a Dice of 0.14, Otsu's bright and dark pixels 0.21 and 0.10; the muscles cover 7.5% of an image on average. TotalSegmentator has no ultrasound task.
  • Tolerance. 1 mm, about 6 pixels: the truth grown or thinned by one pixel keeps a Dice of 0.99 and a surface Dice of 1.0.

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
10 test subjects drawn with seed 2026 out of the 109, then 10 development subjects out of the 99 left. One case a subject: the subject's image of lowest frame number, prone for 106 subjects and standing for the three imaged standing only.
Truth
contours by experts the provider's binary masks, one a muscle, outlined in Fiji by one investigator trained by a spine-imaging expert of over ten years; the two agreed on the cross-sectional area at an ICC of 0.93 to 0.99 on 18 images (Belasso et al., BMC Musculoskeletal Disorders 2020)
Licence
none stated (the database page gives none; the laboratory site's footer reads 'All rights reserved')
10 cases for tuning, never tested

subject_001, subject_070, subject_108, subject_093, subject_006, subject_094, subject_101, subject_089, subject_076, subject_055