Classification

CT Scan Image Preparation and Lung Cancer Classification

Classification

CT Scan Image Preparation and Lung Cancer Classification

MATLAB scripts that process and prepare DICOM files of Lung CT Scan Images into targeted [20x20x20] nodules. A filtering process for CNN training preparation follows this. The CNN scripts are also attached.

The DICOM images of lung cancer for CNN training are obtained from two sources:

• 48 annotated lung patients (C012 - C293 except C113) from LDCT and Projection Data: https://www.cancerimagingarchive.net/collection/ldct-and-projection-data/
• 32 annotated lung patients (R004 - R127) LungCT-Diagnosis: https://www.cancerimagingarchive.net/collection/lungct-diagnosis/

The DICOM images of lung cancer for independent validation are obtained from one source:

• 7 lung patients (R137, R141, R143, R146, R150, R175, R266) LungCT-Diagnosis: https://www.cancerimagingarchive.net/collection/lungct-diagnosis/

The Main script will run ParenchymaSegment, NoduleSearch, and NoduleExtract functions, lung segmentation from DICOM images, searching nodules in the 3D domain, and extracting them into 20 x 20 x 20 dimensions.

NoduleSearch has filtering parameters in the FilterParam function. The cancer nodules are extracted manually from the extracted nodule files and then oversampled using an OverSampling script.

BatchPreps and Train scripts are for deep learning training. There are 5 CNN models available for training:

• Modified U Network (MUNet)
• Modified Double U Network (MDUNet)
• Modified Segmentation Network (MSegNet)
• Modified Deconvolutional Network (MDeConvNet)
• Modified Residual Encoder-Decoder Network (MREDNet)
• Modified Residual Network (MResNet)
• Modified Residual Network with Transformation (MResNeXt)
• Modified Efficient Network (MEffNet)