- Surgical planning by applying the segmentation algorithm Region-Growing.
We are segmenting the spinal disks, by going though the CT volume and putting a seed on each vertebrae.
- Tool calibration by defining the matrices from reference-to-pivot, camera-to-reference, tracker-to-pivot, camera-to-tracker.
- Registration using Iterative Closest Point to align the cloud point for a bone CT scan in blue.
- Artificial Intelligence for CT hip segmentation. We use DICE and Hausdorff score to evaluate the performance. We use a U-NET segmentation for 10 epochs, with different learning rates.
The performance for the hyper-parameter fine-tuning are put in the following table.
| CT | Hip Learning Rate | Epoch | Dice | HD | ASD |
|---|---|---|---|---|---|
| Acetabulum | 0.1 | 5 | 94.96% | 114.97 | 50.38% |
| Femur | 0.1 | 5 | 95.59% | 212.26 | 50.09% |
| Acetabulum | 0.1 | 10 | 92.41% | 117.66 | 66.67% |
| Femur | 0.1 | 10 | 95.40% | 110.00 | 66.49% |
| Acetabulum | 0.001 | 5 | 95.70% | 49.98 | 45.98% |
| Femur | 0.001 | 5 | 96.91% | 129.16 | 44.34% |
| Acetabulum | 0.001 | 10 | 95.99% | 55.00 | 45.01% |
| Femur | 0.001 | 10 | 97.26% | 110.39 | 40.39% |
| Acetabulum | 0.00001 | 5 | 93.92% | 74.92 | 63.94% |
| Femur | 0.00001 | 5 | 94.82% | 111.21 | 68.91% |
| Acetabulum | 0.00001 | 10 | 94.46% | 57.11 | 59.17% |
| Femur | 0.00001 | 10 | 95.83% | 158.23 | 55.78% |