Optimizing computed tomography image reconstruction for focal hepatic lesions: Deep learning image reconstruction vs...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Deep learning image reconstruction (DLIR) is an advanced computed tomography (CT) technique.
• DLIR aims to reduce image noise, improve image quality, and lower radiation doses.
• This study compares DLIR and iterative reconstruction (IR) for evaluating liver lesions.

Purpose of the Study:

• To compare the diagnostic performance of DLIR and IR in assessing focal hepatic lesions.
• To evaluate the impact of different DLIR and IR strengths on image quality and noise.
• To determine the optimal DLIR setting for evaluating hepatic lesions.

Main Methods:

• A retrospective analysis of 216 focal hepatic lesions in 109 adult patients.
• Abdominal CT images were reconstructed using DLIR (low, medium, high) and IR (0-30%).
• Radiologists assessed lesion detectability, borders, confidence, artifacts, and overall quality; image noise was quantitatively measured.

Main Results:

• Significant differences in lesion borders, artifacts, and overall image quality were observed across reconstruction techniques (p < 0.001).
• Low-strength DLIR (DLIR-L) yielded the best overall image quality.
• High-strength DLIR (DLIR-H) minimized noise and artifacts but reduced lesion border quality.

Conclusions:

• Optimal-strength DLIR significantly enhances overall image quality for focal hepatic lesion evaluation compared to IR.
• DLIR-L demonstrated superior overall image quality while maintaining acceptable noise and lesion border characteristics.
• DLIR represents a promising advancement for abdominal CT imaging of hepatic lesions.