SVM isn't classifying its own training images as expected!

I am trying to detect a tumor in a MRI image using fitcsvm. First, I divide every image of my training set into various clusters, then I calculate texture features (using graycomatrix) from each of the clusters that are a part of the tumor. I train the SVM for one-class classification, giving the sets of features calculated for each cluster into fitcsvm, each cluster corresponding to one observation. I do this for several images, but when I decide to test my algorithm with an image that was included in the training set, I get an image which doesn't seem to make sense.

Figure 1 illustrates the MRI training image with the clusters and the Ground Truth (I only train the SVM with the clusters in the MRI image positioned in the dark gray area of the Ground Truth). Figure 2 is the matrix of the scores produced by the predict function when I input the same MRI image to check if the algorithm is able to detect the tumor of one of its own training samples. As I understand it, the pixels in the result image closer to white are the most likely to represent the class for which I trained the algorithm, that is, the tumor.

I also checked if the features calculated for the same cluster in both the training and classifying phase were the same, and they were, so I don't get why the observation isn't correctly classified given that it already integrates the SVM training. I must be able to finish this project, so any help is deeply appreciated!

Figure 1:

Figure 2: