Computer Vision News - January 2017

Results: Input and output are as follows: The training set consisted of 89 pediatric CT scans of both non-TB and TB cases, acquired in 2010 in South Africa and the United Kingdom. The test dataset consisted of 90 pediatric CT scans, 42 non-TB cases and 48 TB cases, acquired in 2012 in South Africa. Accuracy of the classification results are demonstrated in the ROC curves figure below, for two regions of interest: Trachea–RMB–LMB and RMB-RUL-BI. • The LA-PDM on the Trachea–RMB–LMB region achieved an AUC of 0.87 (0.77- 0.94). • The LA-PDM on the RMB-RUL-BI region shows that second order bronchi can also play a role in TB detection with an AUC of 0.81 (0.68-0.90). • As a comparison, a set of features derived from mean airway cross-section obtained an AUC of 0.72 (0.59-0.83) for the Trachea–RMB–LMB region, and an AUC of 0.59 (0.46-0.73) for the RMB-RUL-BI region. Sum up: This paper introduces a method called LA-PDM and its authors claim that the model can accurately distinguish (AUC of 0.81 to 0.87) between airway involvement (pediatric pulmonary TB) and normal airways by examining regions of the airway likely to be affected by lymphadenopathy (Trachea–RMB–LMB and RMB-RUL-BI). This model was trained on a dataset including CT scans of both healthy and involved airways. Its results outperformed a comparison method based on airway cross- sectional features. Computer Vision News Research 35 Research Training Test Input 89 TB bases (with both TB and non-TB cases) The LA-PDMmodel + 90 Unseen CT case (42 non-TB cases and 48 TB cases) Output The LA-PDMmodel TB accuracy prediction The model can accurately distinguish between airway involvement (pediatric pulmonary TB) and normal airways