Performance Evaluation of Three Techniques

Figure 9.44 shows the mean error bar charts for the three pipelines (i.e., using three classification systems: MRF, FCM, and GSM methods).5 The charts can be seen in the Tables 9.1-9.3. Table 9.1 shows the error between the computer-estimated boundary and ground truth boundary using FCM-based

5 We ran the system using each of the three different classifying methods on real patient data. Ground truth boundaries of the walls of the carotid artery were traced for 15 patients. Overall the number of boundary points was roughly 22,500 points. A pixel was equivalent to 0.25 mm. Using MRF, the average error was 0.61 pixels; using FCM, the average error was 0.62 pixels; using GSM, the average error was 0.74 pixels.

Table 9.1: Mean errors as computed using polyline and shortest distance methods when the classification system is FCM based

Patient No.

Artifacted (PDM)

Corrected (PDM)

Artifacted (SDM)

Corrected (SDM)

1

2.052

1.195

2.063

1.216

2

0.928

0.764

0.948

0.794

3

3.174

0.729

3.180

0.756

4

1.106

0.490

1.118

0.513

5

1.514

0.968

1.529

0.993

6

1.079

0.681

1.094

0.704

7

1.278

0.863

1.310

0.893

8

0.928

0.695

0.944

0.723

9

0.758

0.606

0.783

0.631

10

1.004

0.813

1.027

0.840

11

1.407

0.808

1.418

0.826

12

1.408

1.042

1.426

1.078

13

0.735

0.643

0.753

0.670

14

0.922

0.655

0.939

0.685

method. Column 1 shows the error when the estimated boundary is not corrected (artifacted), using the PDM ruler. Column 2 shows the error when the estimated boundary is corrected by merging multiple classes of the lumen, using the PDM ruler. Column 3 shows the error when the estimated boundary is not corrected (artifacted), using the SDM ruler. Column 4 shows the error when the estimated boundary is corrected by merging multiple classes of the lumen, using the SDM ruler. As seen in the table, column 2 shows the least error and is sig-nificiantly improved over the artifacted boundaries. Table 9.2 shows the error between the computer-estimated boundary and ground truth boundary using MRF-based method. Column 1 shows the error when the estimated boundary is not corrected (artifacted), using the PDM ruler. Column 2 shows the error when the estimated boundary is corrected by merging multiple classes of the lumen, using the PDM ruler. Column 3 shows the error when the estimated boundary is not corrected (artifacted), using the SDM ruler. Column 4 shows the error when the estimated boundary is corrected by merging multiple classes of the lumen, using the SDM ruler. As seen in the table, column 2 shows the least error and is significiantly improved over the artifacted boundaries. Table 9.3 shows the error between the computer-estimated boundary and ground truth boundary using GSM-based method. Column 1 shows the error when the estimated boundary is not corrected (artifacted), using the PDM ruler. Column 2 shows the error when the estimated boundary is corrected by merging multiple classes of the lumen,

Table 9.2: Mean errors as computed using polyline and shortest distance methods when the classification system is MRF based

Patient No.

Artifacted (PDM)

Corrected (PDM)

Artifacted (SDM)

Corrected (SDM)

1

1.609

1.SS2

1.627

1.402

2

0.SS1

0.726

0.SB7

0.7B9

3

1.174

0.7S1

1.19B

0.S0B

4

0.6S7

0.BS4

0.706

0.60B

5

l.2S9

0.S9B

1.26S

0.917

6

1.164

1.0S6

1.1S2

1.10B

7

1.100

0.S07

1.124

0.SS9

8

1.004

0.620

1.02S

0.64B

9

0.696

0.679

0.714

0.702

10

0.S90

0.9BS

0.912

0.9S2

11

0.9SS

0.7S6

0.9B4

0.76S

12

0.941

1.06B

0.96B

1.0S9

13

0.679

0.740

0.704

0.761

14

0.SB1

0.694

0.S69

0.716

Table 9.3: Mean errors as computed using polyline and shortest distance methods when the classification system is GSM based

Patient No.

Artifacted (PDM)

Corrected (PDM)

Artifacted (SDM)

Corrected (SDM)

1

1.0S1

1.0S1

1.10B

1.105

2

0.721

0.721

0.746

0.746

S

1.S29

1.S29

1.SB1

1.351

4

0.4S7

0.4S7

0.B0B

0.505

B

0.77S

0.77S

0.S02

0.802

6

0.767

0.767

0.7SS

0.788

7

0.920

0.920

0.949

0.949

S

0.SSB

0.SSB

0.90S

0.903

9

0.BS6

0.BS6

0.BB9

0.559

10

0.S26

0.S26

0.S49

0.849

11

0.7B2

0.7B2

0.774

0.774

12

0.914

0.914

0.942

0.942

13

0.BSS

0.BSS

0.BB7

0.557

14

0.70S

0.70S

0.7S2

0.732

using the PDM ruler. Column 3 shows the error when the estimated boundary is not corrected (artifacted), using the SDM ruler. Column 4 shows the error when the estimated boundary is corrected by merging multiple classes of the lumen, using the SDM ruler. As seen in the table, column 2 shows the least error and is significiantly improved over the artifacted boundaries.

Figure 9.45: Results of estimated boundary using circular- vs. elliptical-based methods. The system used was FCM based. Top rows are circular-based ROI, while the corresponding bottom rows are elliptical-based ROIs.
0 0

Post a comment