| Abstract: |  Lung nodules are the lung parenchyma structures found by radiodiagnostic imaging methods, especially Computed Tomography. Lung nodules are of various etiologies and can be found in various lung diseases . At worst they represent a primary or secondary tumorous proces of the lung. That is why it is necessary to find all suspicious lung nodules.
The aim of this study is to create the automatic lung nodule detection algorithm, based on the existing one. In my work I first analyse the baseline algorithm results to find all the shortcomings that can be improved to receive better output results. These findings are applied to create new classification method. This metod is based on reducing the number of existing nodule characteristics, modifying the training data and applying the suitable classifier to receive as good sensitivity and as low number of false positive detections as possible. For that purpose, combinations of several dimensionality reduction methods and several classifiers are studied.
New method have the same sensitivity, but significantly lower number of false positives than the existing one.
|
|---|