ETD

• Artificial Intelligence
• Computed Tomography
• Deep Learning Image Reconstruction
• Image Quality
• Iterative Reconstruction
• Liver Metastases

The accurate detection of liver metastases from diverse primary malignancies is
fundamental for oncological staging. Computed Tomography (CT) remains the gold standard,
but traditional image reconstruction techniques face a trade-off between radiation dose and
spatial resolution. Analytical methods suffer from excessive noise, while Iterative
Reconstruction (IR) algorithms often apply non-linear smoothing that obscures delicate lesion
margins.
To address these limitations, this thesis quantitatively and qualitatively evaluates the
clinical impact of Deep Learning Image Reconstruction (DLIR) compared to standard
Adaptive Statistical Iterative Reconstruction (ASiR). The methodology incorporates a dual

approach: an objective analysis measuring Signal-to-Noise Ratio (SNR) and Contrast-to-
Noise Ratio (CNR), and a subjective, blinded evaluation conducted by radiology residents.

Utilizing a 5-point Likert scale, the readers assessed images reconstructed at different levels
to determine their overall preference and diagnostic quality.
By systematically comparing ASiR and DLIR, this study highlights how artificial
intelligence influences both measurable physical metrics and subjective clinical perception.
The integrated findings determine the optimal reconstruction strategy to preserve natural noise
texture, maximize lesion conspicuity, and ultimately improve diagnostic confidence in hepatic
oncology.