Using A Mix Of Radiomics And Machine Learning, Accurate Prediction Of Pure Uric Acid Urinary Stones In A Clinical Setting

Goal: Urinary acid stones can be effectively and non-invasively treated with oral chemolysis. The objective of this study was to use non-contrast-enhanced computed tomography scans (NCCTs) to categorize urinary stones into two groups: pure uric acid (pUA) and other composition (Others).

Methods: We examined instances that were managed at our institution between 2019 and 2021. Based on composition analyses, they were classified as pUA or Others and randomly divided into training and testing data sets. Several cases had several NCCTs, all of which were gathered. Every urinary stone was handled as a separate sample in every NCCT. We retrieved both original and wavelet radiomics features for each sample from manually created volumes of interest. The Least Absolute Shrinkage and Selection Operator was then used to choose the most crucial features, which were then used to construct the final model using a Support Vector Machine. Accuracy, sensitivity, specificity, and area under the precision-recall curve (AUPRC) were used to assess performance on the testing set.

Results: A total of 576 samples were obtained from 302 cases, 118 of which had pUA urinary stones. Ten most significant characteristics were ultimately chosen from 851 original and wavelet radiomics features that were retrieved for each sample. Accuracy, sensitivity, specificity, and AUPRC on the testing data set were 90.8%, 100%, 87.5%, and 0.902 for per-instance prediction and 93.9%, 97.9%, 92.2%, and 0.958 for per-sample prediction.

In conclusion, pUA urinary stones can be reliably predicted by the machine learning system that was trained using radiomics features from NCCTs. Our research points to a possible aid for choosing a stone disease treatment.