Low contrast detectability performance of model observers based on CT phantom images: kVp influence

Area of Science:

• Medical Physics
• Radiological Imaging
• Image Analysis

Background:

• Low contrast detectability (LCD) is crucial for interpreting CT images.
• Assessing human performance in LCD tasks is vital for clinical applications.
• Model observers offer a potential tool for evaluating and optimizing CT imaging protocols.

Purpose of the Study:

• To compare the LCD performance of two model observers (NPWE and CHO) against human observers in CT phantom images.
• To investigate the effect of varying kilovoltage peak (kVp) levels on LCD.
• To determine the suitability of model observers for predicting human performance and aiding CT protocol optimization.

Main Methods:

• CT phantom images were acquired at different kVp levels (80-135 kV) and object contrasts (0.5-1%).
• Human observers performed a 2-alternative forced choice (2-AFC) experiment to assess object visibility.
• Two model observers, Non-Prewhitening Matched Filter with an eye filter (NPWE) and Channelized Hotelling Observer (CHO), were used to analyze LCD.

Main Results:

• kVp significantly affected object contrast, with variations up to 17%.
• Both models initially overestimated human performance but were corrected using efficiency and internal noise factors.
• The NPWE model demonstrated superior agreement with human performance (Pearson's r ≥ 0.976) compared to CHO (Pearson's r ≥ 0.706).
• Bland-Altman analysis showed better agreement between NPWE and human results (Δ=-0.3%, Δ±2σ=[-4.0%,4.5%]) than CHO (Δ=-1.2%, Δ±2σ=[-10.7%,8.3%]).

Conclusions:

• The NPWE model observer shows significant potential for accurately predicting human performance in CT low contrast detection tasks.
• This model can be a valuable tool for optimizing CT imaging protocols, particularly in kVp selection.
• Further validation is recommended for broader application in clinical settings.