Table 3 Detailed BRAF status classification performance metrics for the model pretrained using semiautomatic whole tumor volume segmentations, radiomics features, and tumor location

From: Segmentation-free pretherapeutic assessment of BRAF-status in pediatric low-grade gliomas

  1. Throughout Table 3, values highlighted in green/red illustrate where the model is particularly strong (green) or weak (red). The first column of Table 3a describes the four metrics we used to evaluate the model. Each value in the table represents the mean of each metric, calculated across the 25 test sets. Accuracy was defined as the percentage of patients overall for which the true label was predicted by the model and was not calculated on a per-class basis. All other metrics were calculated according to their standard definition, for each class, and then those per-class values were averaged to obtain the values listed in the overall column. The second column of Table 3a details the values of the performance metrics overall. The remaining columns detail the performance metrics on a per-class basis. Table 3b contains the confusion matrix for the classification model, where the primary value is the mean of the number of patients in the cell across 25 trials, with the percentage of patients in parentheses. It can be seen from the confusion matrix that the model excels at identifying BRAF Fusion, but frequently mistakes non-BRAF altered tumors for BRAF mutations.
Back to article page