ABSTRACT
PURPOSE: Since semi-automated lesion quantification may be more precise than manual uni- and bidimensional measurements, the purpose of this study was to compare semi-automated with manual evaluations of cervical, thoracic and abdominal lymph nodes in patients with malignant lymphoma. MATERIALS AND METHODS: 62 patients with known malignant lymphoma underwent staging with contrast-enhanced 16-MDCT (16x0.7 mm coll., 120 kV, cervical/thoracic/abdominal: 150/120/160 mAseff., 1/1.25 /1 pitch, 4/3 - 5/4 - 5/4 slice thickness/reconstruction increment). On the basis of these standard reconstructed slices, each lesion was quantified in terms of RECIST and its longest orthographic diameter using a semi-automated software tool (Syngo CT Oncology, Siemens Medical Solutions, Forchheim, Germany) and manually by an experienced radiologist. The degree of agreement between manual measurements and software quantification was statistically assessed by computing the concordance correlation coefficient kappa and represented graphically in corresponding Bland-Altman plots. RESULTS: 74/80 cervical, 51/80 thoracic and 75/80 abdominal lymph nodes were correctly evaluated by the software. A strong degree of agreement between both measurement techniques (RECIST diameter: kappa = 0.97 (cervical)/0.98 (thoracic)/0.99 (abdominal); longest orthographic diameter: kappa = 0.97/0.93/0.97) was obtained. CONCLUSION: Semi-automated measurement of cervical, thoracic and abdominal lymph nodes showed valid results on standard axial reconstructions compared to manual quantification with the limitation of a high false segmentation rate in thoracic lymph nodes.
Subject(s)
Abdominal Neoplasms/diagnostic imaging , Artificial Intelligence , Head and Neck Neoplasms/diagnostic imaging , Image Interpretation, Computer-Assisted/methods , Lymphography/methods , Lymphoma, Non-Hodgkin/diagnostic imaging , Radiography, Dual-Energy Scanned Projection/methods , Software , Thoracic Neoplasms/diagnostic imaging , Tomography, Spiral Computed/methods , Whole Body Imaging/methods , Adult , Aged , Aged, 80 and over , Algorithms , Contrast Media/administration & dosage , Female , Humans , Image Interpretation, Computer-Assisted/standards , Lymphography/standards , Male , Middle Aged , Radiography, Dual-Energy Scanned Projection/standards , Radiology Information Systems , Reference Values , Retrospective Studies , Sensitivity and Specificity , Tomography, Spiral Computed/standards , Whole Body Imaging/standards , World Health OrganizationABSTRACT
PURPOSE: To assess the accuracy of liver lesion measurement using automated measurement and segmentation software depending on the vascularization level. MATERIALS AND METHODS: Arterial and portal venous phase multislice CT (MSCT) was performed for 58 patients. 94 liver lesions were evaluated and classified according to vascularity (hypervascular: 13 hepatocellular carcinomas, 20 hemangiomas; hypovascular: 31 metastases, 3 lymphomas, 4 abscesses; liquid: 23 cysts). The RECIST diameter and volume were obtained using automated measurement and segmentation software and compared to corresponding measurements derived visually by two experienced radiologists as a reference standard. Statistical analysis was performed using the Wilcoxon test and concordance correlation coefficients. RESULTS: Automated measurements revealed no significant difference between the arterial and portal venous phase in hypovascular (mean RECIST diameter: 31.4 vs. 30.2 mm; p = 0.65; kappa = 0.875) and liquid lesions (20.4 vs. 20.1 mm; p = 0.1; kappa = 0.996). The RECIST diameter and volume of hypervascular lesions were significantly underestimated in the portal venous phase as compared to the arterial phase (30.3 vs. 26.9 mm, p = 0.007, kappa = 0.834; 10.7 vs. 7.9 ml, p = 0.0045, kappa = 0.752). Automated measurements for hypovascular and liquid lesions in the arterial and portal venous phase were concordant to the reference standard. Hypervascular lesion measurements were in line with the reference standard for the arterial phase (30.3 vs. 32.2 mm, p = 0.66, kappa = 0.754), but revealed a significant difference for the portal venous phase (26.9 vs. 32.1 mm; p = 0.041; kappa = 0.606). CONCLUSION: Automated measurement and segmentation software provides accurate and reliable determination of the RECIST diameter and volume in hypovascular and liquid liver lesions. Hypervascular lesions are prone to be underestimated with regard to size in the portal venous phase and therefore should preferentially be segmented in the arterial phase.