Contribution of an artificial intelligence deep-learning reconstruction algorithm for dose optimization in lumbar...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Iterative reconstruction algorithms are standard for lumbar spine CT.
• Assessing novel reconstruction techniques is crucial for dose reduction and image quality.

Purpose of the Study:

• To evaluate the impact of a new artificial intelligence deep-learning reconstruction (AI-DLR) algorithm on image quality and radiation dose.
• To compare AI-DLR with the standard iterative reconstruction algorithm (iDose4) in lumbar spine CT.

Main Methods:

• Phantoms were scanned using a tube current modulation system at varying DoseRight Indexes (DRI).
• Raw data were reconstructed using iDose4 (Level 4) and AI-DLR (Smoother, Smooth, Standard) with a bone kernel.
• Image quality metrics including Noise Power Spectrum (NPS), task-based transfer function (TTF), and detectability index (d') were computed. Subjective assessment by radiologists was also performed.

Main Results:

• AI-DLR demonstrated lower noise magnitude compared to iDose4, with noise decreasing from Standard to Smoother levels.
• Spatial frequency of NPS was similar between iDose4 and Standard AI-DLR, but decreased with Smoother AI-DLR.
• Detectability index (d') for simulated lesions increased with AI-DLR from Standard to Smoother levels, showing higher detectability at lower DRI.
• Radiologists rated images obtained with lower DRI and AI-DLR (Smooth/Smoother) as satisfactory for clinical use.

Conclusions:

• The AI-DLR algorithm, particularly at Smooth and Smoother levels, allows for significant radiation dose reduction (up to 72%) in lumbar spine CT.
• This dose reduction is achieved while maintaining high detectability of lytic and sclerotic bone lesions.
• The AI-DLR algorithm provides clinically acceptable overall image quality, offering a promising alternative to iterative reconstruction for lumbar spine CT examinations.