Classification of urinary stones by cluster analysis of ionic composition data

Area of Science:

• Urology
• Computational Chemistry
• Data Science

Background:

• Kidney stones are classified using empirical rules based on urate, oxalate, and phosphate concentrations.
• A dataset of 214 kidney stones was previously classified using these empirical rules.
• The effectiveness of cluster analysis for kidney stone classification needs further investigation.

Purpose of the Study:

• To classify kidney stones using cluster analysis with nine chemical parameters.
• To compare cluster analysis with empirical classification rules.
• To explore alternative effective classifiers and evaluate different clustering methods, distance measures, and standardization techniques.

Main Methods:

• Cluster analysis was applied to data from 214 kidney stones.
• Nine chemical analysis parameters were utilized.
• Various clustering techniques, distance/similarity measures, and data standardization methods were compared.
• Results were benchmarked against a prior empirical classification.

Main Results:

• Including six additional parameters beyond urate, oxalate, and phosphate did not enhance classification accuracy.
• The best classification match (6% error) was achieved using average linkage (between groups) clustering with squared Euclidean distance, without data standardization.
• Excluding the three main radicals resulted in a 63% matching error.
• Cluster analysis indicated that carbon ions alone could serve as a single classifier with approximately 10% error.

Conclusions:

• Comprehensive chemical analysis data does not necessarily improve kidney stone classification accuracy over simpler methods.
• Specific clustering parameters (average linkage, squared Euclidean distance) and minimal data standardization yield optimal results.
• Carbon ion concentration alone shows promise as a single, effective classifier for kidney stones.