figure a

Dear Editor,

We have read this article, published in your journal this year (Li, H, Xie, Y, Liu, H. et al. Non-Contrast CT-Based Radiomics Score for Predicting Hematoma Enlargement in Spontaneous Intracerebral Hemorrhage. Clin Neuroradiol 32, 517–528 (2022). https://doi.org/10.1007/s00062-021-01062-w). There are a few ideas that I would like to share with you.

Firstly, the authors did not describe exactly the number of hematoma expansion positive cases in their results.

Secondly, in the results section, when they presented the performance of the predictive model, the values did not match the figures, for example, the description “The AUC under the ROCs of radiomics model in training cohort was 0.82 (0.72–0.93), and the sensitivity and specificity were 0.42 (0.36–0.49) and 0.92 (0.87–0.97), respectively (Fig. 4a). The constructed radiomics model from the training cohort was introduced to test cohort, and the AUC, sensitivity and specificity of validated dataset were 0.83 (0.71–0.95), 0.33 (0.27–0.40) and 0.88 (0.79–0.97), respectively (Fig. 4b).” in the results section, while in the figure 4a, AUC is 0.85(0.78–0.91) in the train cohort. In addition, the values of sensitivity and specificity in the training and test cohorts cannot be derived from the optimal threshold points of the AUC curves in Fig. 4.

In the figure 4, the best cutoff value obtained from ROC curves in the training cohort.

Yours sincerely,

Kai Chen