Lymph node task card

To train the system and assess the performance of the automated lymph node detection, reference lymph node annotations were created for each MRL. The annotation comprised a consensus reading by two expert readers: an MD researcher specially trained in reading MRL scans (2 years of MRL experience, >300 MRLs), and an abdominal radiologist (>10 years of MRL experience, >1000 MRLs).
The T1-weighted (VIBE) sequence, which is insensitive to USPIO contrast, was used for localization and assessment of shape and size of the lymph nodes. In the VIBE images, the individual lymph nodes were interactively segmented using the application Lymph Node Task Card, developed by Siemens, Malvern, PA (USA). The T2*-weighted (MEDIC) sequence was used to assess USPIO uptake for the clinical diagnosis of the patient but was not used in our analysis. A total of 5,089 lymph nodes were annotated, 21 per patient on average. All lymph nodes visible in the images were annotated; none were excluded based on small size or otherwise.

The contrast uptake in all detected lymph nodes was computed as follows. All lymph nodes visible in the pelvic region were interactively segmented using the computer application Lymph Node Task Card (Siemens, Malvern, PA). Segmentation was performed based on the T1-weighted images, in which all lymph nodes (normal as well as metastatic ones) appear as hypointense structures. From these three-dimensional segmentations, the volume of each lymph node was recorded automatically.
As a measure of contrast uptake, relative SI was computed rather than absolute SI. Lymph node assessment based on relative SI (local fat calibrated lymph node assessment) is more similar to visual assessment of the MRL image by a radiologist: visual assessment is also based on a comparison of the SI of a lymph node with the SI of the surrounding fat. Local fat calibration compensates for coil profiles and other factors creating a spatially varying SI distortion in the images. The fat calibration was implemented by manually segmenting a region of fatty tissue in the direct neighborhood of each lymph node. The relative SI is calculated by subtraction of the mean SI of the local fatty tissue region from the SI in the corresponding lymph node.