Please enable JavaScript.
Coggle requires JavaScript to display documents.
Kidney Stones, Note: Lacked necessary metrics for comparison in previous…
Kidney Stones
Identification of kidney stone types
Necessary to determine appropriate treatments
Prevention of further patient complications
Standard identification
Use morphoconstitutional analysis
Time-consuming procedure
Tedious extraction
Difficult to identify even for experts
Methods of identification
Previous works
Use ex-vivo images
Highly controlled setting
Achieved accuracy as high as 94%
Paper contirbution
Use of endoscopic in-vivo data
Faster than standard procedures
Analysis of four kidney stone compositions
Analysis of surface and cross-section images
Dataset
Necessary for deep learning methods
Clinical Image Dataset
90 surface images
87 cross-section images
Increased number of samples
Patch extraction
Technique in which a segmented kidney stone images is placed in a grid
Used 256x256 overlapping pixel patches
Achieved the highest score from experiments
Created a total of 2680 surface images
Increased cross-section images to 2470
Data augmentation
Patch flipping
Perspective distortions
Affine transformations
Whitening
Handling class imbalance
Necessary to prevent overfitting
Used up-sampling approach
Adds new patches not located in grid
Sampling from original image
Experiment and Results
Used five machine learning models
AlexNet
VGG19
XGBoost
Inception
Random Forest
XGBoost and Inception achieved highest precision and recall scores
Note: Lacked necessary metrics for comparison in previous works i.e. accuracy scores
Overall: The results could not be compared to ex-vivo applications. It lacked necessary metrics for further analysis. Some statements were also different from results. They noted that inception is the best for all types, when in fact XGBoost yielded better scores for Uric Acid. Still the paper is a novel approach and could hasten diagnosis.
Lopez, F., Varela, A., Hinojosa, O., Mendez, M., Trinh, D.-H., ElBeze, J., Hubert, J., Estrade, V., Gonzalez, M., Ochoa, G., and Daul, C. Assessing deep learning methods for the identification of kidney stones in endoscopic images, 2021.