vPSNR: a visualization-aware image fidelity metric tailored for diagnostic imaging

Area of Science:

• Medical Imaging
• Data Compression
• Image Fidelity

Background:

• Diagnostic imaging generates vast data, straining IT systems and radiologists.
• Effective data reduction is needed beyond lossless compression to manage data overload.
• Existing image fidelity metrics lack support for high-dimensional data, visualization awareness, and data independence.

Purpose of the Study:

• To introduce a novel image fidelity metric, visual peak signal-to-noise ratio (vPSNR).
• To address limitations of current metrics in modern clinical settings.
• To ensure metrics support high-dimensional data, visualization awareness, and independence from original data.

Main Methods:

• Developed and tested the visual peak signal-to-noise ratio (vPSNR) metric.
• Conducted image fidelity tests on CT datasets.
• Evaluated the impact of visualization transforms on diagnostic quality using an observer-based study.

Main Results:

• The visualization transform significantly impacts diagnostic quality.
• vPSNR effectively captures the effect of visualization transforms on image fidelity.
• Tests confirmed the broad applicability of vPSNR across various data reduction scenarios.

Conclusions:

• vPSNR addresses a critical gap in existing image fidelity metrics for clinical applications.
• While not a sole solution, vPSNR serves as a valuable complement for image fidelity assessment.
• The metric is relevant for diverse data reduction scenarios in medical imaging.