Automatic Hardy and Clapham’s classification of hallux sesamoid position on foot radiographs using deep neural network
Description
We developed the deep neural network model for the automatic classification of hallux sesamoid position according to Hardy and Clapham classificaion. This dataset includes the trained neural network model, the code for the development of neural network, and the validation process and statistic analysis. The article corresponds to this dataset is "Automatic Hardy and Clapham’s classification of hallux sesamoid position on foot radiographs using deep neural network" .
Files
Steps to reproduce
'Data' includes the trained neural network model as 'trainedNet.mat'. The model was developed in MATLAB 2023a. Other file in 'Data' folder includes the prediction results by the developed and the manual classifications by three orthopedic surgeons. This dataset has no radiograph because of the protection of the patient's privacy. To reproduce the prediction, execute 'TryWholeProcess.mat' in 'Prediction' folder. The code outputs the Hardy and Clapham's classification and the prediction images when you input the DICOM file including unilateral foot radiograph. 'Inter_AI_Surgeon' folder includes the code for the calculation of weighted Kappa value between DNN model and the median of three surgeons. 'InterRater' folder calculates the weighted Kappa between 1-2 , 2-3, 3-1, surgeons. 'IntraRater_session1and2' calculates the weighted Kappa between 1st and 2nd sessoin of each surgeon. 'Statistical subfunction' has the code to calculate the weighted Kappa and the comparison of the weighted Kappa values usind z-test. This subfunction partially contains the code from https://github.com/dnafinder/Cohen (see licence.txt)