Overview / Usage

A Computer-aided Diagnosis System that can detect early-stage lung cancer with higher accuracy than expert radiologists.
The project's objectives are:

• Affordable system (enable people to do early-stage detection at very cheap costs comparing to the traditional methodologies)
• High accuracy
• Very fast diagnosis (the result will be out as soon as the X-Ray CT Scan enters the system)
• Visualization

Methodology / Approach

We are using multiple Deep Learning CNNs to detect the cancerous nodules which are the abnormal cells that indicate the development or the existanace of lung cancer. These cells are the primary objects whom human radiologists look for in the CT Scan.
The System has five main phases:
1- Preprocessing: remove the noise from the CT Scan, unify spacing between the 3D-CT Scans's 2D slices.
2- Lung Segmentation: segment the lung precisley from the entire CT Scan in order to reduce the search space.
3- Nodule Detection and Segmentation: detect the nodules within the segmented lung 3D image, if a nodule is found, a 3D cubic patch is taken around it and fed to the next phases.
4- Nodule Classification: not all nodules detected are true nodules, i.e. blood vessels, the system needs to classify a nodule into two classes: either malignant or benign.
4- False Positives Reduction: reduce the occurrences of false positives and false negatives by optimizing the system in terms of its accuracy, sensitivity and the overall performance (precession, recall metrics).

Technologies Used

• 3D Convolutional Neural Networks, using Tensor-flow Deep Learning framework (Optimized version by Intel) -- python

• OpenCV -- python

• SimpleITK -- python

• Scipy -- python

• Scikit -- python

• LUNG16 dataset