Automatic Detection, Segmentation and Classification of Pulmonar Nodules in CT Images



LNDetector aims at the development of a Computer-aided Diagnosis (CAD) system for the detection, segmentation and classification of nodular volumes in chest Computer Tomography (CT) scans.
The main output is a lung CAD prototype to be evaluated in the Radiology Department of FMUP, Porto.


Main Task Goals:

Ground truth dataset

To create a ground truth (GT) dataset, with radiologists annotations on lung nodule location and characterization, together with the radiologist attention chart mapping the radiologist’s attention during a diagnosis process.

Nodule detection

To automatically detect the nodule locations within the lung volume, particularly for the more challenging nodule locations, e.g. juxtapleural or juxtavascular and different texture solidities, e.g. ground glass opacities (GGOs).

Nodule segmentation and characterization

To automatically compute the boundary surface of each detected nodule.

Nodule classification

To grade the lung nodules in a malignancy scale.