Automatische Detektion und volumetrische Segmentierung der Milz in CT-Untersuchungen

Zusammenfassung

Ziel: Die automatische Detektion und Volumetrie der Milz in CT-Untersuchungen durch die THESEUS-MEDICO-Software. Evaluation der Übereinstimmung der Ergebnisse von automatischer (aV), geschätzter (gV) und manueller (mV) Milzvolumetrie auch hinsichtlich der Veränderung des Milzvolumens zwischen 2 Untersuchungen.

Material und Methoden: Evaluation der auf Verfahren des „marginal space learning“ und auf „boosting-Algorithmen“ beruhenden CAD-Software anhand von 3 konsekutiven CT-Untersuchungen (Thorax/Abdomen; portalvenöse Kontrastmittelphase; 1 bzw. 5 mm Schichtdicke) von 15 konsekutiven Lymphompatienten. Bestimmung der gV: 30 cm³ + 0,58 (Breite × Tiefe × Höhe der Milz). Die Vermessung der Milz und die als Referenzstandard dienende mV erfolgte durch erfahrene Radiologen.

Ergebnisse: Die aV konnte in allen CT-Untersuchungen innerhalb von 15,2 (± 2,4) s durchgeführt werden. Das durch die aV ermittelte Milzvolumen betrug durchschnittlich 268,21 ± 114,67 cm³, das der mV 281,58 ± 130,21 cm³ und das der gV 268,93 ± 104,60 cm³. Der Korrelationskoeffizient zwischen aV und mV betrug 0,99 (Determinationskoeffizient (R²) = 0,98), zwischen mV und gV 0,91 (R² = 0,83) und zwischen aV und gV 0,91 (R² = 0,82). Es zeigte sich eine Korrelation von 0,92 (R² = 0,84) zwischen aV und mV bei der Veränderung des Milzvolumens zwischen 2 Untersuchungszeitpunkten in konsekutiven CT-Untersuchungen. Zwischen mV und gV wurde ein Korrelationskoeffizient von 0,95 (R² = 0,91) und zwischen aV und gV von 0,83 (R² = 0,69) ermittelt.

Schlussfolgerung: Die automatische Detektions- und Segmentierungs-Software ermöglicht eine schnelle und genaue Volumetrie der Milz in CT-Untersuchungen. Die ohne Mehraufwand bereitgestellte Information über das Milzvolumen und die Veränderung des Milzvolumens zwischen 2 Untersuchungen ist bei unterschiedlichen klinischen Fragestellungen relevant.

Abstract

Purpose: To introduce automated detection and volumetric segmentation of the spleen in spiral CT scans with the THESEUS-MEDICO software. The consistency between automated volumetry (aV), estimated volume determination (eV) and manual volume segmentation (mV) was evaluated.

Materials and Methods: Retrospective evaluation of the CAD system based on methods like “marginal space learning” and “boosting algorithms”. 3 consecutive spiral CT scans (thoraco-abdominal; portal-venous contrast agent phase; 1 or 5 mm slice thickness) of 15 consecutive lymphoma patients were included. The eV: 30 cm³ + 0.58 (width × length × thickness of the spleen) and the mV as the reference standard were determined by an experienced radiologist.

Results: The aV could be performed in all CT scans within 15.2 (± 2.4) seconds. The average splenic volume measured by aV was 268.21 ± 114.67 cm³ compared to 281.58 ± 130.21 cm³ in mV and 268.93 ± 104.60 cm³ in eV. The correlation coefficient was 0.99 (coefficient of determination (R²) = 0.98) for aV and mV, 0.91 (R² = 0.83) for mV and eV and 0.91 (R² = 0.82) for aV and eV. There was an almost perfect correlation of the changes in splenic volume measured with the new aV and mV (0.92; R² = 0.84), mV and eV (0.95; R² = 0.91) and aV and eV (0.83; R² = 0.69) between two time points.

Conclusion: The automated detection and volumetric segmentation software rapidly provides an accurate measurement of the splenic volume in CT scans. Knowledge about splenic volume and its change between two examinations provides valuable clinical information without effort for the radiologist.

Key words

CAD - splenic volumetry - CT

• Literatur

• 1 Strijk SP, Wagener DJ, Bogman MJ et al. The spleen in Hodgkin disease: diagnostic value of CT. Radiology 1985; 154: 753-757
• 2 Hancock SL, Scidmore NS, Hopkins KL et al. Computed tomography assessment of splenic size as a predictor of splenic weight and disease in involvementin laparotomy staged Hodgkin‘s disease. Int J Rad Oncol Biol Phys 1994; 28: 93-99
• 3 Thurn P, Bücheler E, Lackner KJ et al. Einführung in die radiologische Diagnostik. Stuttgart: Thieme; 1998: 458-460
• 4 Heymsfield SB, Fulenwider T, Nordlinger B et al. Accurate measurement of liver, kidney, and spleen volume and mass by computerized axial tomography. Ann Intern Med 1979; 90: 185-187
• 5 Prassopoulos P, Daskalogiannaki M, Raissaki M et al. Determination of normal splenic volume on computed tomography in relation to age, gender and body habitus. Eur Radiol 1997; 7: 246-248
• 6 Revel MP, Bissery A, Bienvenu M et al. Are Two-dimensional CT Measurements of Small Noncalcified Pulmonary Nodules Reliable?. Radiology 2004; 231: 453-458
• 7 Erasmus JJ, Gladish GW, Broemeling L et al. Interobserver and intraobserver variability in measurement of non-small-cell carcinoma lung lesions: implications for assessment of tumor response. J Clin Oncol 2003; 21: 2574-2582
• 8 Thiesse P, Ollivier L, Di Stefano-Louineau D et al. Response rate accuracy in oncology trials: reasons for interobserver variability. Groupe Francais d’Immunotherapie of the Federation Nationale des Centres de Lutte Contre le Cancer. J Clin Oncol 1997; 15: 3507-3514
• 9 Zillner S, Huber M. Intelligent Medical Image Search. IT Information Technology 2009; 51: 285-293
• 10 Seifert S, Barbu A, Zhou S et al. Hierarchical parsing and semantic navigation of full body CT data. SPIE Medical Imaging; 2009
• 11 Tu Z, Zhou XS, Barbu A et al. “Probabilistic 3D Polyp Detection in CT Images: The Role of Sample Alignment”. IEEE CVPR; 2006: 1544-1551
• 12 Tu Z. “Probabilistic Boosting-Tree: Learning Discriminative Models for Classification, Recognition, and Clustering”. 10th IEEE ICCV; 2005 1589–1596
• 13 Oren M, Papageorgiou C, Sinha P et al. “Pedestrian detection using wavelet templates”. IEEE CVPR; 1997: 193-199
• 14 Schapire R, Singer Y. “Improved Boosting Algorithms Using Confidence-rated Predictions”. Machine Learning 1999; 37: 297-336
• 15 Zheng Y, Barbu A, Georgescu B et al. “Fast Automatic Heart Chamber Segmentation from 3D CT Data Using Marginal Space Learning and Steerable Features”. ICCV; 2007
• 16 Cootes TF, Hill A, Taylor CJ. The use of active shape models for locating structures in medical images. Image and Vision Computing 1994; 12: 355-366
• 17 Sofka M, Zhang J, Zhou SK et al. Multiple object detection bysequential Monte Carlo and hierarchical detection network. IEEE CVPR; 2010
• 18 Cootes TF, Taylor CJ. Active Shape Models – „Smart Snakes“. BMVC92. Proceedings of the British Machine Vision Conference. Berlin: Springer-Verlag; 1992: 266-275
• 19 Brenner DJ, Hall EJ. Computed tomography--an increasing source of radiation exposure. N Engl J Med 2007; 357: 2277-2284
• 20 Andriole KP, Morin RL, Arenson RL et al. Addressing the coming radiology crisis-the Society for Computer Applications in Radiology transforming the radiological interpretation process (TRIP) initiative. J Digit Imaging 2004; 17: 235-243
• 21 Andriole KP, Wolfe JM, Khorasani R et al. Optimizing analysis, visualization, and navigation of large image data sets: one 5000-section CT scan can ruin your whole day. Radiology 2011; 259: 346-362
• 22 Li WX, Zhao XT, Chai WM et al. Hepatitis B virus-induced liver fibrosis and cirrhosis: the value of liver and spleen volumetry with multi-detector spiral computed tomography. J Dig Dis 2010; 11: 215-223
• 23 Chen TY, Chen CL, Huang TL et al. Does hepatic graft weight affect the reduction of spleen size after living donor liver transplantation?. Transplant Proc 2010; 42: 882-883
• 24 Ishifuro M, Horiguchi J, Ohshita A et al. Remission of splenic volume in the recipient during the course of adult-to-adult living donor liver transplantation: evaluation using multislice computed tomography and semiautomatic software. Transplant Proc 2009; 41: 3772-3774
• 25 Ibrahim S, Chen CL, Wang CC et al. Liver regeneration and splenic enlargement in donors after living-donor liver transplantation. World J Surg 2005; 29: 1658-1666
• 26 Hidaka H, Nakazawa T, Wang G et al. Reliability and validity of splenic volume measurement by 3-D ultrasound. Hepatol Res 2010; 40: 979-988
• 27 Farraher SW, Jara H, Chang KJ et al. Liver and spleen volumetry with quantitative MR imaging and dual-space clustering segmentation. Radiology 2005; 237: 322-328
• 28 Heitmann KR, Rückert S, Heussel CP et al. Automatische Berechnung des Milzvolumens aus Spiral-CT-Daten mit Hilfe neuronaler Netze und “Fuzzy Logik”. Fortschr Röntgenstr 2000; 172: 139-146
• 29 Linguraru MG, Sandberg JK, Li Z et al. Atlas-based Automated Segmentation of Spleen and Liver using Adaptive Enhancement Estimation. Med Image Comput Comput Assist Interv 2009; 5762: 1001-1008