Figure 2.

Experimental protocol. Due to the small sample size, the LOOCV method was used to estimate the prediction performance. In each iteration, one sample was held out for test and the remaining samples were used for training. The regularization parameter of a logistical regression model was estimated through 10-fold cross-validation using the training data, and then a predictive model was trained using the estimated parameter and blindly applied to the held-out sample. The experiment was repeated until each sample had been tested.