For technical descriptions of our method, please refer to the following paper:
Hongming Xu, Sunho Park, Tae Hyun Hwang, "Computerized Classification of Prostate Cancer Gleason Scores from Whole Slide Images", TCBB 2019. (https://ieeexplore.ieee.org/document/8836110)
In the following, the steps to run our programs are provided.
The folder "TCGA_PRAD_Gleason_Score_Prediction_CSLBP" includes Matlab programs of our method, which has four subfolders:
- step1)feature_extraction
- step2)feature_classification
- step3)plot_figures
- step4)comparison
The folder "tcga_prad_tcbb_dl" mainly includes Python programs for us to test baseline deep learning models for comparison in the paper
Once your environment has met the requirments (see requirements), you could run our method by following steps:
-
(1) go to the folder "step1)feature_extrction", then run the function "main_feasExtraction.m"
-
To run the "main_feasExtrcation.m", make sure that YOU should change the imageTrainPath (line 39) and imageDebugPath (line 43) to the local address of your computer
-
The extracted image features are finnally saved into the .mat files, please feel free to check and revise in lines: 161-162, for your convenience
Overall, by running the main_feasExtraction.m, we descrbe each WSI by a 1152 dimensional feature vector, i.e.,
-
-
(2) go to the folder "step2)feature_classification", you could run our classifcation evaluations
-
In the folder tcga_288_25, it includes the .mat files with features computed by running the function "main_feasExtraction.m" on our dataset.
-
So if you want to see our results, you could directly run the function "main_crossValiation.m". NOTE that you could change the lines 51-54 to select SVM classifier with different kernels
-
If you computed features on your dataset with our function "main_feasExtraction.m", you need to adaptively make changes such as the line 11 in main_crossValidation.m for cross-validation testing.
-
-
(3) the folder "step3)plot_figures" includes the matlab files for generating figures, you could run them directly
-
(4) the folder "step4)comparion" includes textrue features for comparison, you could run them similarly as "step2)feature_classification"