FIGURE 6.

Results from the predictive models for patients from the test set. Four biomarkers were used for predictions (diffusion‐weighted imaging [DWI], mean transit time [MTT], time to peak [TTP], and Clinical data). After deep convolutional neural network (CNN_deep), the cutoff value of the DMTC model in predicting the hemorrhagic transformation (HT) is 0.0009681931. (A) Patient A is a 61‐year‐old man (NIHSS = 15), scanned 3.5 h after symptom onset. The predicted value is 0.0107461931, which means the patient having HT after therapy. Follow‐up CT at 24 h showed HT, which is consistent with the model prediction. (B) Patient B is a 58‐year‐old man (NIHSS = 18), scanned 3 h after symptom onset. The predicted value is 0.00017365169, which means the patient having no HT after therapy. Follow‐up CT at 24 h showed no HT, which is consistent with the model prediction. (C) Patient C is a 60‐year‐old man (NIHSS = 13), scanned 3.5 h after symptom onset. The predicted value is 0.00051431544, which means the patient having no HT after therapy while follow‐up CT at 24 h showed HT. (D) Patient D is a 65‐year‐old man (NIHSS = 16), scanned 4 h after symptom onset. The predicted value is 0.001586112, which means the patient having HT after therapy while follow‐up CT at 24 h showed no HT. Patients C and D showed that the prediction model may give erroneous results [Colour figure can be viewed at wileyonlinelibrary.com]