Image quality and dose reduction opportunity of deep learning image reconstruction algorithm for CT: a phantom study

Area of Science:

• Medical imaging
• Radiology
• Artificial intelligence in healthcare

Background:

• Iterative reconstruction (IR) algorithms have improved CT image quality and reduced radiation dose.
• Deep learning image reconstruction (DLIR) is an emerging technique with the potential for further advancements.

Purpose of the Study:

• To evaluate the impact of a novel DLIR algorithm on CT image quality and radiation dose reduction.
• To compare the performance of DLIR against a hybrid IR algorithm.

Main Methods:

• A standard phantom was scanned at seven dose levels.
• Raw data were reconstructed using filtered back projection (FBP), hybrid IR (ASiR-V), and three levels of DLIR (TrueFidelity™).
• Image quality was assessed using noise power spectrum (NPS), task-based transfer function (TTF), and detectability index (d') for simulated lesions.

Main Results:

• DLIR reduced noise and increased spatial frequencies compared to IR.
• DLIR improved task-based transfer function (TTF) at all levels.
• Detectability (d') was higher with DLIR, especially for low-contrast lesions, outperforming hybrid IR at higher dose levels.

Conclusions:

• The new DLIR algorithm effectively reduces noise and enhances spatial resolution and detectability.
• DLIR offers significant potential for radiation dose optimization in CT imaging.
• DLIR provides improved image quality without altering noise texture, unlike some IR methods.