Academia.eduAcademia.edu

Figure 5 – uploaded by Bartjan de Hoop

See full PDFdownloadDownload figure
Fig. 4. Typical results of the automatic segment segmentation for 3 (the columns) randomly selected scans on different cross-sectional slices. Slices have been chosen in such a way that all segments are shown in each direction at least once.

Figure 4 Typical results of the automatic segment segmentation for 3 (the columns) randomly selected scans on different cross-sectional slices. Slices have been chosen in such a way that all segments are shown in each direction at least once.

Related Figures (9)

Fig. 1. Schematic drawing of the bronchial tree up to segmental level. Based on the anatomy of the bronchial tree 10 segments are defined in the right lung and eight in the left lung.
Fig. 1. Schematic drawing of the bronchial tree up to segmental level. Based on the anatomy of the bronchial tree 10 segments are defined in the right lung and eight in the left lung.
Fig. 2. Illustration of pulmonary lobar and segmental anatomy from actual results of the different steps of the segment segmentation algorithm on a randomly selected scan. From left to right: an axial slice, a coronal slice, a sagittal slice of the right lung and a sagittal slice of the left lung. The top row shows slices of the original scan. The second row shows the result of the lobe segmentation, in the bottom row the corresponding segment segmentation is shown.
Fig. 2. Illustration of pulmonary lobar and segmental anatomy from actual results of the different steps of the segment segmentation algorithm on a randomly selected scan. From left to right: an axial slice, a coronal slice, a sagittal slice of the right lung and a sagittal slice of the left lung. The top row shows slices of the original scan. The second row shows the result of the lobe segmentation, in the bottom row the corresponding segment segmentation is shown.
Fig. 3. Example of the fissure segmentation, (a) the original CT slice, (b) result of supervised enhancement, (c) the final fissure segmentation.
Fig. 3. Example of the fissure segmentation, (a) the original CT slice, (b) result of supervised enhancement, (c) the final fissure segmentation.
on the same scanners and reconstructed with the same kernel as the data from the screening trial described above. In these scans a human observer manually indicated the lobar fissures in every fourth coronal slice. Segmenting fissures was performed by clicking points on the fissure; between two points, a straight line was automatically drawn.
on the same scanners and reconstructed with the same kernel as the data from the screening trial described above. In these scans a human observer manually indicated the lobar fissures in every fourth coronal slice. Segmenting fissures was performed by clicking points on the fissure; between two points, a straight line was automatically drawn.
CONFUSION MATRICES FOR THE SEGMENT SEGMENTATION IN EACH LUNG. NUMBERS ARE PERCENTAGES. THE COLUMNS INDICATE THE CLASSIFICATION, THE ROWS INDICATE THE GROUND TRUTH. SO, FOR EXAMPLE ENTRANCE (0, 1) = 2 INDICATES THAT OF ALL FINDINGS THAT SHOULD BE, ACCORDING TO THE REFERENCE ANNOTATIONS, IN SEGMENT 1, 2% ARE CLASSIFIED AS BEING IN SEGMENT 2 10 was mainly confused with segment 6. Note that all those pairs of segments are neighboring. In addition, in the right lung seg- ment 2 was often confused for segment | and 3 (15% and 11%, respectively) by the automatic system. This is comparable to the interobserver agreement (15% and 13%, respectively). fused with segment 3 in the specific lungs. In the right lung this was probably due to incompleteness or absence of the minor fissure separating the upper and middle lobe. Since there is usually no minor fissure in the left lung it is even more dif- ficult to assign the correct segmental labels to positions near segmental borders. The fact that different segments were con- fused with segment 3 in each lung is due to the different ori- entation of segment 4 and 5 in both lungs, as can be seen from Fig. 4.
CONFUSION MATRICES FOR THE SEGMENT SEGMENTATION IN EACH LUNG. NUMBERS ARE PERCENTAGES. THE COLUMNS INDICATE THE CLASSIFICATION, THE ROWS INDICATE THE GROUND TRUTH. SO, FOR EXAMPLE ENTRANCE (0, 1) = 2 INDICATES THAT OF ALL FINDINGS THAT SHOULD BE, ACCORDING TO THE REFERENCE ANNOTATIONS, IN SEGMENT 1, 2% ARE CLASSIFIED AS BEING IN SEGMENT 2 10 was mainly confused with segment 6. Note that all those pairs of segments are neighboring. In addition, in the right lung seg- ment 2 was often confused for segment | and 3 (15% and 11%, respectively) by the automatic system. This is comparable to the interobserver agreement (15% and 13%, respectively). fused with segment 3 in the specific lungs. In the right lung this was probably due to incompleteness or absence of the minor fissure separating the upper and middle lobe. Since there is usually no minor fissure in the left lung it is even more dif- ficult to assign the correct segmental labels to positions near segmental borders. The fact that different segments were con- fused with segment 3 in each lung is due to the different ori- entation of segment 4 and 5 in both lungs, as can be seen from Fig. 4.
CONFUSION MATRICES FOR THE LOBE SEGMENTATION IN EACH LUNG, THE NUMBERS ARE PERCENTAGES. THE COLUMNS INDICATE THE CLASSIFICATION. THE ROWS INDICATE THE GROUND TRUTH. 1 INDICATES THE UPPER LOBE, 2 INDICATES THE MIDDLE LOBE AND 3 THE LOWER LOBE
CONFUSION MATRICES FOR THE LOBE SEGMENTATION IN EACH LUNG, THE NUMBERS ARE PERCENTAGES. THE COLUMNS INDICATE THE CLASSIFICATION. THE ROWS INDICATE THE GROUND TRUTH. 1 INDICATES THE UPPER LOBE, 2 INDICATES THE MIDDLE LOBE AND 3 THE LOWER LOBE
Fig. 5. Example of PFOs and the corresponding lobe segmentation as produced by the automatic lobe segmentation. The red cross indicates the point marked in the nodule.
Fig. 5. Example of PFOs and the corresponding lobe segmentation as produced by the automatic lobe segmentation. The red cross indicates the point marked in the nodule.
Fig.6. Example result of the segment segmentation on a scan containing patho- logic abnormalities. Despite the presence of abnormalities the result of the seg- ment segmentation is comparable to the results on the screening data.
Fig.6. Example result of the segment segmentation on a scan containing patho- logic abnormalities. Despite the presence of abnormalities the result of the seg- ment segmentation is comparable to the results on the screening data.
Related topics:
EngineeringImage ProcessingMedical ImagingComputed TomographyRegion growingObserver VariationAutomatic ClassificationX ray Computed Tomography
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved