Identification of Prostate Cancer Cell Nuclei for DNA-Grading of Malignancy

Abstract

DNA Image Cytometry is a method for early cancer diagnosis and grading of cancer, using a photomicroscopic system to measure the DNA content of nuclei. Specifically for the prostate, this method can be used to distinguish between clinically insignificant, non-aggressive tumors, and those which need to be removed or irradiated. This decision is based on the analysis of the DNA distribution among examined nuclei. However, even trained personnel usually requires more than 40 minutes for collecting the requested number of nuclei. Considering a shortage of skilled personnel, reducing the interaction time with the system is desired. Towards this end, a training set consisting of 47982 Feulgen stained nuclei and features mainly based on the nucleus morphology are used to train a Random Forest classifier. A motorized microscope was used to automatically scan ten slides from a test set and classify their nuclei. Using the leaving one out strategy, the classifier achieved a classification rate of 90.93% on the training set. For the test set, the resulting DNA distribution of each measurement was evaluated by a pathological expert. The DNA grades of malignancy of the automated measurement were identical to the grades of the corresponding manual reference measurements in all cases. Interaction time required for grading was reduced to approximately five minutes per case for manually validating the classified nuclei in diagnostically relevant DNA ranges.