Improving computer aided disease detection using knowledge of disease appearance.

The accurate quantification of disease patterns in medical images allows radiologists to track the progress of a disease. Various computer vision techniques are able to automatically detect different patterns that appear on images. However, classical pattern detection approaches do not perform satisfactorily on medical images. The problem is that texture descriptors, alone, do not capture information that is pertinent to medical images, i.e. the disease appearance and distribution. We present a method that uses knowledge of anatomy and specialised knowledge about disease appearance to improve computer-aided detection. The system has been tested on detecting honeycombing - a diffuse lung disease pattern in HRCT images of the lung. The results show that the proposed knowledge guided approach improves the accuracy of honeycombing detection. A paired t-test, shows the improvement in accuracy to be statistically significant (p<0.0001).