Automated quantitative and qualitative analysis of whole neuroblastoma tumour images for prognosis

Manual quantitative and qualitative microscopic analysis of cancerous tumours is subject to inter-intra observer variability in pathology. Neuroblastoma is an infant cancer with one of the lowest survival rates. Choosing a proper therapeutic regime for the tumour is highly dependent on determining the tumour aggressiveness level which requires an extensive microscopic analysis. There is an urgent demand from pathologists for reducing the role of microscopic analysis in the process of prognosis and using an automated system to determine the tumour aggressiveness. In this paper, we develop an automated system to address this demand. We propose a novel fourstage hybrid algorithm. First, we develop novel whole slide image partitioning and zooming techniques. Second, we introduce an image enhancement technique to reduce the intensity variation within the tissue images. Third, we deploy a thresholding technique for segmenting the regions of interest. Fourth, we develop a prognosis decision making engine based on a robust clinical prognosis scheme to classify the aggressiveness level using the segmented regions of interest. The performance of the system is evaluated by a pathologist. The system is compared against a state-of-the-art system, and the results indicate a superiority for our system in grading the tumour with average F-measure 86.77%.