Skip to content

A repository for the analysis and segmentation of lesion annotations and CT Scan slices, respectively.

Notifications You must be signed in to change notification settings

Dashora7/DeepLesion-Segmentation-with-TensorFlow-Python

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

52 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DeepLesion-Segmentation-with-TensorFlow-Python

A repository for the analysis of CT Scan slices and lesion annotations. Through this code, lesion mask labels can be created and adjusted, TF Record data can be produced, and Mask Region-based Architectures can be utilized (Inception-Resnet V2 was used).

This code is meant to be used with DeepLesion data and TensorFlow Object Detection API in Python. Download the data with the download script and cite the DeepLesion research.

Train_test_split can be used to split the available CSV data into two data sets. To split the images, use image_split.

Image Resizer is set to resize to 512x512 to ensure stable dimensionality for CNN flow.

The Mask Tester File can be used to create mask annotations through statistical analysis of image pixels. By using 3D Mask Propagation, we can also interpolate and produce intermediate slice annotations through a volume.

A supplementary Otsu's Algorithm Analysis is also provided. This is not what we used for mask generation though.

The generate_tf_records_new file is used to take the data produced by 3D Propagation and convert it into data readable by TensorFlow Object Detection API.

Inspection of these Records can be done through the inspect_records file.

After producing this data, it can be fed into TensorFlow ODAPI. We have supplied some basic files for the usage of ODAPI in the TF running folder. Also, model hyperparameters are in pipeline.config

About

A repository for the analysis and segmentation of lesion annotations and CT Scan slices, respectively.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published