Deep learning reconstruction for improving the visualization of acute brain infarct on computed tomography

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Acute infarct detection is critical for timely stroke management.
• Traditional CT reconstruction methods may have limitations in visualizing subtle infarcts.
• Advancements in AI, such as deep learning reconstruction (DLR), offer potential improvements in image quality and diagnostic accuracy.

Purpose of the Study:

• To compare the effectiveness of deep learning reconstruction (DLR) versus hybrid iterative reconstruction (Hybrid IR) for depicting acute infarcts on head CT.
• To evaluate the impact of DLR on image quality and noise reduction in the context of acute stroke imaging.

Main Methods:

• Retrospective analysis of unenhanced head CT images from patients with and without acute infarction.
• Reconstruction of CT images using both DLR and Hybrid IR techniques.
• Qualitative assessment of infarct conspicuity and image quality by three readers.
• Quantitative analysis of image noise using regions of interest in specific brain structures.

Main Results:

• DLR demonstrated superior conspicuity of acute infarcts compared to Hybrid IR, with statistically significant differences for two readers.
• The conspicuity of infarcts within 24 hours of symptom onset was significantly improved with DLR for all readers.
• DLR resulted in a significant reduction in both qualitative and quantitative image noise compared to Hybrid IR.

Conclusions:

• Deep learning reconstruction (DLR) enhances the depiction of acute infarcts on head CT.
• DLR is particularly beneficial for imaging acute infarcts within the first 24 hours after symptom onset.
• The use of DLR leads to improved image quality and reduced noise in head CT examinations for stroke evaluation.