The Expectation-Maximization (EM) algorithms are iterative methods that find the maximum likelihood estimated parameters over a statistical model. Each algorithm iteration is performed in two steps - The Expectation step (E-step) which measures the log-likelihood expectation on some already estimated parameter. Then a Maximization step (M-step) maximizes the expected log-likelihood calculated in the E-step.
| EM Segmentation Result | EM Data distribution |
|---|---|
 |
 |
| Patient | CSF-Kmeans | GM-Kmeans | WM-Kmeans | CSF-EM | GM-EM | WM-EM | Time (s) | Iterations |
|---|---|---|---|---|---|---|---|---|
| Case 1 | 81.56% | 69.63% | 82.30% | 85.04% | 82.22% | 90.67% | 19.71 | 44 |
| Case 2 | 74.64% | 84.79% | 69.48% | 68.79% | 83.74% | 15.83% | 28.73 | 59 |
| Case 3 | 71.35% | 83.80% | 81.42% | 77.95% | 83.73% | 85.56% | 18.83 | 41 |
| Case 4 | 77.80% | 82.13% | 68.83% | 86.90% | 90.08% | 84.00% | 17.67 | 40 |
| Case 5 | 83.68% | 79.17% | 85.43% | 88.99% | 84.64% | 86.57% | 22.58 | 47 |