A comparison study assessing the feasibility of ultrasound-initialized deformable bone models

Figures & data

Figure 1. The first two eigen modes of variation of our proximal femur model. The shape instances were generated by evaluating with .

Figure 2. (a) The DRB attached to the ultrasound probe. (b) Automatically segmented bone contours in anatomical space.

Table I.  Error statistics of predicted shapes for Bone 1.

Bone 1	Trial number	Pointer-based error (mm)		Ultrasound-based error (mm)
		Mean	Median	Mean	Median
User 1	1	2.20	1.71	14.34	14.34
	2	1.99	1.46	9.52	8.91
	3	1.98	1.25	7.01	6.69
	4	2.06	1.63	7.97	7.18
	5	1.79	1.56	11.94	10.68
	6	1.91	1.65	N/A	N/A
	Average	1.99	1.54	10.15	9.56
User 2	1	2.24	1.81	3.88	3.56
	2	2.00	1.74	4.37	4.28
	3	1.76	1.59	6.88	6.34
	4	2.11	1.85	4.75	4.51
	5	2.21	1.90	3.08	2.53
	Average	2.06	1.78	4.59	4.35

Table II.  Error statistics of predicted shapes for Bone 2.

Bone 2	Trial Number	Pointer-based error (mm)		Ultrasound-based error (mm)
		Mean	Median	Mean	Median
User 1	1	2.95	2.78	5.63	5.72
	2	2.78	2.62	4.19	4.33
	3	2.77	2.59	2.90	2.58
	4	2.91	2.53	6.54	6.35
	5	2.56	2.39	3.48	3.51
	Average	2.80	2.58	4.55	4.50
User 2	1	2.83	2.68	2.94	2.12
	2	3.66	3.45	5.30	4.98
	3	3.06	2.96	3.79	3.48
	4	3.30	3.17	4.57	4.54
	5	3.59	3.43	3.12	2.84
	Average	3.29	3.14	3.95	3.59

Figure 3. Statistics accumulated from the trials carried out by the two users on (a) Bone 1 and (b) Bone 2 using a tracked pointer to obtain surface points. The average mean error is plotted against the number of surface points digitized.

Figure 4. For each population, the average mean error is plotted against the number of digitized points.

Table III.  Mean surface errors for nine dry cadaver bones in the CT-based error scheme.

Cadaver bone number	Mean error (mm) w.r.t. number of points
	Large population			Small population
	3	27	54	3	27	54
1	2.08	1.9	1.72	2.57	2.02	1.85
2	0.96	0.91	0.85	2.03	1.49	1.23
3	2.44	2.28	2	3.02	2.69	2.5
4	2.55	2.45	2.03	2.92	2.63	2.12
5	2.18	1.99	1.85	1.98	1.87	1.72
6	3.49	3.1	2.54	4.44	3.79	2.65
7	1.73	1.59	1.39	3.15	2.61	2.23
8	2.01	1.87	1.67	1.91	1.75	1.58
9	2.06	2.04	1.83	2.22	2.14	1.64
Average	2.17	2.01	1.76	2.69	2.33	1.95

Figure 5. Pointer-based prediction: predicted models overlaid onto ‘gold’ references. Bone 1 (left): mean error = 1.76 mm; and Bone 2 (right): mean error = 2.78 mm.

Figure 6. Ultrasound-based prediction: predicted models overlaid onto ‘gold’ references. Bone 1 (left): mean error = 3.08 mm; and Bone 2 (right): mean error = 2.90 mm.