📄️ No Classification
If no classification is selected, the image is not classified and thus, the labels that have been drawn in the image are not changed but used directly in later post processing and information calculation.
📄️ Deep learning
Deep learning and convolutional neural networks are a broader family of AI machine learning methods. It involves neural network algorithms that use a cascade of many layers of nonlinear processing units for feature extraction and transformation with each successive layer using the output from the previous layer as input. Using deep learning for classification allows you to segment abstract image structures that would be impossible to segment with a simple pixel classifier.
📄️ Threshold
Using Threshold classification, it is possible to specify the upper and lower threshold values for each label. When specifying the threshold of several features, it is possible to obtain a more specific classification.
📄️ Bayesian
This method can be used if there is heavy variation throughout the data set. Under Type, choose between Linear and Quadratic classification. Linear classification discriminates between classes using straight lines. Quadratic classification uses 2nd degree polynomials for the discrimination.
📄️ Decision forest
The decision forest classifier can be tuned by adjusting the slider between Fast and Accurate. The values on the slider can be between 100 and 1, where 100 signifies a more advanced classifier 1 signifies a simpler classifier. It is often not necessary to use the most advanced decision forest to get satisfactory classification results.
📄️ K-means clustering
K-means Clustering is a very useful method to compensate for variation in Staining and Lighting throughout the dataset. To use K-Means clustering, the individual feature bands must be comparable (i.e. they must be approximately the same scale).
📄️ Detect Region Of Interest
Detect Region of Interest is a simple method that divides the image into sections, based on pixel intensity. This method is a useful tool for separation of the background and tissue.
📄️ Phenotype
The phenotyping algorithm uses a two-layered clustering algorithm using the euclidean distance in the band space as clustering measure. The first clustering layer roughly separates/clusters cells into main groups where further separation in subgroups (phenotypes) is provided by the second layer clustering. For instance, the algorithm may cluster immune cells into the same main group with two sub-phenotypes for cytotoxic T-cells and helper T-cells.
📄️ Cell classification
Cell Classification is a useful tool for detecting individual cells. Ensure that a number of image classes for training have been created and labelled according to cell types in the images. Recall that more training labels, means that more data will be used to train the classifier. Thus, drawing many training labels, leads to a better performing classifier. Also recall that all image classes defined need to have drawn training labels of its type.