AJR: Semiautomatic lymph node measurement more reliable
Cancerous lymph nodes measured semiautomatically using computer software yielded greater interobserver precision than nodes measured manually, according to research published in the October issue of the American Journal of Roentgenology.
Manual measurement of lymph nodes “is not only time-consuming but also bears inherent sources of error, evidenced by high interobserver and intraobserver variability,” author Boris Buerke, MD, of University of Muenster, Germany, and colleagues wrote.
The authors also noted that, unlike the semiautomatic measurements taken in their study, manually acquired node measurements “are greatly influenced by the orientation of the lymph node within the body … In oncologic follow-up examinations, interobserver bias becomes crucial when tumor response to chemotherapy is graded on the basis of manually obtained metric criteria.”
Two radiologists reviewed images of 742 cancerous lymph nodes in 112 patients using a multi-detector CT. All nodes had diameters measuring at least 4 mm for their larger axes and were either peripheral (cervical, axillary or inguinal), thoracic or abdominal lesions.
The diameters of the long-axes and short-axes of all nodes were measured, as well as the volumes of all nodes, using both manual and semiautomatic techniques. Manual measurement was taken by electronic caliper, while for semiautomatic measurements radiologists simply traced the length of the short- and long-axes. Algorithmic software used the marks to calculate both the diameters and volumes of nodes.
All semiautomated measurements were either approved or corrected by both readers. Sixty-five percent of semiautomated measurements required no correction by radiologists, the authors reported. Thoracic nodes and smaller nodes (nodes with larger axes measuring less than 15 mm) required the most corrections, with 55 percent and 45 correction rates, respectively.
Researchers evaluated variations in measurements between radiologists for both manual and semiautomatic techniques by looking at the correlation and standard deviations of the inter-radiologist measurements. The authors found that semiautomatic measurements varied significantly less between readers than manual measurements. The correlations between the radiologists’ semiautomatic and manual measurements of nodes were .81 and .7, respectively. The standard deviation between radiologists for short-axis nodes was 2.7 percent lower for semiautomated than manual measurements (20.9 percent versus 23.6 for manual measurements). These findings were statistically significant.
The authors also found that the variability in radiologists’ semiautomatic measurements of node volumes was especially low, with an average difference of only 3 mm. The authors emphasized the importance of this statistic in relation to cancer treatment, as consistent, precise volumetric measurements between radiologists are crucial for monitoring the progression of patients’ lesions through chemotherapy.
Buerke and colleagues noted that, unlike previous studies, which only measured node size in phantoms (in vitro), the present in vivo study differed from past studies in that larger, not smaller, nodes were measured with the greatest precision.
Despite the relative accuracy of semiautomated measurements in the study, the authors observed that the semiautomatic method systematically over-measured nodes by an average of 11 to 14 percent of node size. Buerke and colleagues hypothesized that this over-measurement was partly the result of the semiautomatic software incorporating the partial volume effect into the node measurements, while typically this mass would not be included on manually measured nodes.
The authors concluded: “Semiautomatic lymph node measurements, particularly lymph node volume, are—irrespective of lymph node location—precise in terms of reproducibility and appear to be considerably more reliable than manual lymph node assessments.”
Manual measurement of lymph nodes “is not only time-consuming but also bears inherent sources of error, evidenced by high interobserver and intraobserver variability,” author Boris Buerke, MD, of University of Muenster, Germany, and colleagues wrote.
The authors also noted that, unlike the semiautomatic measurements taken in their study, manually acquired node measurements “are greatly influenced by the orientation of the lymph node within the body … In oncologic follow-up examinations, interobserver bias becomes crucial when tumor response to chemotherapy is graded on the basis of manually obtained metric criteria.”
Two radiologists reviewed images of 742 cancerous lymph nodes in 112 patients using a multi-detector CT. All nodes had diameters measuring at least 4 mm for their larger axes and were either peripheral (cervical, axillary or inguinal), thoracic or abdominal lesions.
The diameters of the long-axes and short-axes of all nodes were measured, as well as the volumes of all nodes, using both manual and semiautomatic techniques. Manual measurement was taken by electronic caliper, while for semiautomatic measurements radiologists simply traced the length of the short- and long-axes. Algorithmic software used the marks to calculate both the diameters and volumes of nodes.
All semiautomated measurements were either approved or corrected by both readers. Sixty-five percent of semiautomated measurements required no correction by radiologists, the authors reported. Thoracic nodes and smaller nodes (nodes with larger axes measuring less than 15 mm) required the most corrections, with 55 percent and 45 correction rates, respectively.
Researchers evaluated variations in measurements between radiologists for both manual and semiautomatic techniques by looking at the correlation and standard deviations of the inter-radiologist measurements. The authors found that semiautomatic measurements varied significantly less between readers than manual measurements. The correlations between the radiologists’ semiautomatic and manual measurements of nodes were .81 and .7, respectively. The standard deviation between radiologists for short-axis nodes was 2.7 percent lower for semiautomated than manual measurements (20.9 percent versus 23.6 for manual measurements). These findings were statistically significant.
The authors also found that the variability in radiologists’ semiautomatic measurements of node volumes was especially low, with an average difference of only 3 mm. The authors emphasized the importance of this statistic in relation to cancer treatment, as consistent, precise volumetric measurements between radiologists are crucial for monitoring the progression of patients’ lesions through chemotherapy.
Buerke and colleagues noted that, unlike previous studies, which only measured node size in phantoms (in vitro), the present in vivo study differed from past studies in that larger, not smaller, nodes were measured with the greatest precision.
Despite the relative accuracy of semiautomated measurements in the study, the authors observed that the semiautomatic method systematically over-measured nodes by an average of 11 to 14 percent of node size. Buerke and colleagues hypothesized that this over-measurement was partly the result of the semiautomatic software incorporating the partial volume effect into the node measurements, while typically this mass would not be included on manually measured nodes.
The authors concluded: “Semiautomatic lymph node measurements, particularly lymph node volume, are—irrespective of lymph node location—precise in terms of reproducibility and appear to be considerably more reliable than manual lymph node assessments.”