IEEE Transactions on Pattern Analysis and Machine Intelligence
Correlation-based Feature Selection for Discrete and Numeric Class Machine Learning
ICML '00 Proceedings of the Seventeenth International Conference on Machine Learning
Data Mining: Practical Machine Learning Tools and Techniques, Second Edition (Morgan Kaufmann Series in Data Management Systems)
Hi-index | 0.00 |
Honeycombing is a disease pattern seen in High-Resolution Computed Tomography which allows a confident diagnosis of a number of diseases involving fibrosis of the lung. An accurate quantification of honeycombing allows radiologists to determine the progress of the disease process. Previous techniques commonly applied a classifier over the whole lung image to detect lung pathologies. This resulted in spurious classifications of honeycombing in regions where the presence of honeycombing was highly improbable. In this paper, we present a novel technique which uses a seeded region growing algorithm to guide the classifier to regions with potential honeycombing. We show that the proposed technique improves the accuracy of the honeycombing detection. The technique was tested using ten-fold cross validation on forty two images over eight different patients. The proposed technique classified regions of interests with an accuracy of 89.7%, sensitivity of 96.6% and a specificity of 88.6%.