Low-contrast detectability and potential for radiation dose reduction using deep learning image reconstruction-A...

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• A novel deep learning image reconstruction (DLIR) algorithm for computed tomography (CT) has recently received clinical approval.
• DLIR represents a significant advancement in CT image processing technology.

Purpose of the Study:

• To evaluate the low-contrast detectability and radiation dose reduction capabilities of CT images reconstructed using DLIR.
• To compare the performance of DLIR against traditional filtered back projection (FBP) and hybrid iterative reconstruction (IR) methods.

Main Methods:

• A CT phantom with simulated liver lesions was scanned at various dose levels (5-25 mGy).
• Images were reconstructed using FBP, hybrid IR (IR50), and three strengths of DLIR (DLL, DLM, DLH).
• Low-contrast lesion detectability was assessed by 20 readers; dose reduction potential was modeled.

Main Results:

• DLIR significantly improved low-contrast detectability compared to FBP and IR50, particularly at lower dose levels (5 and 10 mGy).
• For example, DLH improved detectability by up to 12.3 percentage points versus FBP at 10 mGy.
• Estimated dose reduction potential relative to FBP was 39% for DLM and 55% for DLH; relative to IR50, it was 21% for DLM and 42% for DLH.

Conclusions:

• The DLIR algorithm enhances low-contrast detectability in CT imaging.
• DLIR demonstrates considerable potential for reducing patient radiation dose while maintaining or improving image quality.