This Article Abstract Full Text Full Text (PDF) Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Wiemker, R Articles by Fleiter, T Search for Related Content PubMed PubMed Citation Articles by Wiemker, R Articles by Fleiter, T

Aspects of computer-aided detection (CAD) and volumetry of pulmonary nodules using multislice CT

¹ Philips Research Laboratories Hamburg, Germany, ² Department of Radiology, Charité Hospital, Humboldt University Berlin, Germany, ³ Philips Medical Systems CT, Haifa, Israel, ⁴ Philips Medical Systems CT Clinical Science, Cleveland, OH, USA, ⁵ Philips Medical Systems Medical IT, Best, The Netherlands and ⁶ University Hospitals of Cleveland, Ohio, USA

View larger version (146K): [in a new window]   Figure 1. Example of a central lung nodule (4 mm diameter) hidden in the maze of the pulmonary vessel tree (maximum intensity projection).

View larger version (188K): [in a new window]   Figure 2. Vessels and airways of the lung, visualised after automated segmentation of a high resolution CT data set (1 mm slice thickness, 400 slices, inverted maximum intensity projection).

View larger version (181K): [in a new window]   Figure 3. Computer-detected nodule candidates (here a large nodule and a more subtle one, which is also three-dimensionally rendered in Figure 8) are marked on the axial slice images.

View larger version (77K): [in a new window]   Figure 4. Overview of the computer-detected nodules indicated in a coronal view of the thorax CT data set.

View larger version (101K): [in a new window]   Figure 5. Graphical user interface for inspection of the computer-detected nodules. A mouse click on a nodule yields a close-up window with a rotatable slab maximum intensity projection and the volume rendering of the automatically segmented nodule (bottom right).

View larger version (144K): [in a new window]   Figure 6. Maximum projection renderings of a variety of pulmonary nodules of different sizes (ordered by volume from top left to bottom right). The wide spectrum of possible sizes is one of the challenges that had to be mastered for computer detection algorithms.

View larger version (68K): [in a new window]   Figure 7. Volume rendering of a juxta-pleural nodule with attached vessels.

View larger version (74K): [in a new window]   Figure 8. Volume rendering of a central nodule (4 mm) with attached vessels.

View larger version (27K): [in a new window]   Figure 9. Example of a single nodule as it appears above different Hounsfield thresholds: the vascular connectivity increases with decreasing Hounsfield threshold (surface renderings, CT data: 0.8 mm slice spacing, 1 mm slice thickness).

View larger version (29K): [in a new window]   Figure 10. Example of a single nodule as it appears in different slice thicknesses (simulated from original data of 1 mm slice thickness): the vascular connectivity decreases with increasing slice thickness (surface renderings at a threshold of –500 HU).

View larger version (35K): [in a new window]   Figure 11. Top: Example of a true nodule attached to the lung wall. Bottom: Sketch of such a nodule to illustrate the segmentation algorithm; left: start point is set, not necessarily at the centre of the mass; right: the areas that are iteratively added to the region expansion, growing first into lobular areas, but then eventually also into the lung wall and connected vessels. The final cut-off surface is denoted by a solid black line.

View larger version (61K): [in a new window]   Figure 12. Example of automated matching of a nodule between follow-up examinations reconstructed with different imaging protocol parameters.