RETURN TO ABSTRACT LISTING


Objective:
The performance of deep learning models for the detection of intracranial hemorrhage depends on many factors. We aimed to investigate the impact of patient's risk factors on the accuracy of an FDA-approved, commercially available, deep learning algorithm for the detection of intracranial hemorrhage.

Materials and Methods:
This prospective, IRB-approved study included all patients (> 18 years) who underwent CT imaging from different clinical settings (i.e., emergency room, inpatient, and outpatient) in our institute from July 2020–February 2021. The risk factors of the patients have been collected from the patients’ EMR, such as history of hypertension, diabetes, smoking, trauma, and intracranial operative intervention. Univariate and multivariate analysis for the risk factors have been conducted to view its impact on software performance. To analyze the data, we used Microsoft Excel and IBM SPSS v28.

Results:
There were 5600 (2823 [50.41%] women and 2777 [49.59%] men) patients included in this study. The mean age at presentation was 57.97 ± 19.66 years (range 18–104 years). In the univariate analysis, there was a statistically significant correlation between the software results and age, history of trauma, intracranial operative intervention, and patient’s history of hypertension. In the multivariate analysis, we found a statistically significant correlation for patient’s age, history of trauma, and intracranial operative intervention.

Conclusion:
The performance of deep learning-based models can be influenced by the patient’s risk factors. Interpretation of deep learning models for the detection of intracranial hemorrhage might be influenced by patient’s risk factors.