Computers tackle lymph node volume measurement

As the incidence of malignant melanoma increases, researchers are seeking ways to accurately and reproducibly measure tumor volume and therapy response. Computer-aided volumetric analysis may be an answer.

Semiautomated software program accurately measures lymph node volume in half the time needed to do so manually. (Provided by M. Fabel-Schulte)

Dr. Michael Fabel-Schulte and colleagues from the radiology department at the German Cancer Research Center in Heidelberg tested 3D semiautomated segmentation and volumetry of lymph nodes in 25 patients with malignant metastatic melanoma (OncoTreat, MeVis, Germany). CT scanning covered the neck, chest, abdomen, and pelvis. They reported the study at the 2006 European Congress of Radiology.

Two independent readers evaluated 120 suspicious lymph nodes by using volume, time, segmentation quality, and number of corrections. Additionally, 20 lymph nodes were segmented manually.

The software performance allowed a reliable volume analysis that was faster than manual segmentation, Fabel-Schulte said. Segmentation quality was rated acceptable to excellent in 81% of lymph nodes by reader one and 79% by reader two. Correlation of the volume was highly significant between both readers. Readers had to manually correct 15% of the lymph nodes.

The average time for automated segmentation per lymph nodes was 70 to 100 seconds, compared with 180 to 200 seconds for the primary manual segmentation. Encouraged by the results, researchers suggest further study in a larger patient population.

This software tool is not in clinical use and needs to be further validated, Fabel-Schulte said. It is, to his knowledge, the only system of its kind for volumetric analysis of lymph nodes.

"Most existent and implemented software tools are for volumetric analysis of lung nodules. Lymph node measurement is far more challenging," he said.

He suggested the software would be commercially available in 2007.

In another study, investigators tested the software's ability to reliably reproduce volume measurement analysis of lung and liver tumors.

Led by Lars Bornemann, Ph.D., a computer scientist with MeVis, researchers tested the system with 96 lung metastases, which ranged in size from 4.6 mm to 60 mm. It showed a low interobserver variability of the volume analysis (median: 0.1%) and a low interscan variability (median: 4.7%).

A second reproducibility study with 86 liver metastases clocked a median interobserver variability of 7%. The average processing time for each lesion was about two seconds.

Bornemann concluded that the proposed software tool has the potential to improve oncological treatment planning and monitoring with respect to accuracy, robustness, speed, and convenience.

The software is composed of hybrid segmentation algorithms that make use of contrasts in density as well as of morphology information to segment the different types of lesions.

For follow-up exams, the system locates the corresponding lesion in the new scan by automatically matching corresponding lesion positions. All information about the processed lesions is summarized in a report, which includes a table containing volumes, volume growth, and doubling time for each lesion.

For more information from the Diagnostic Imaging archives:

CAD helps inexperienced readers with CT colonography

Advocates urge uniformity for change analysis tools

Symposium boosts concept of 'interventional oncology'

Lung CAD can avert malpractice lawsuits