Deep learning-based CT slice synthesis improves radiomic feature reproducibility and discriminative performance in...

Area of Science:

• Medical imaging analysis
• Radiomics and artificial intelligence
• Computational pathology

Background:

• CT slice thickness significantly impacts radiomic feature (RF) reproducibility and discriminative power.
• Thick-slice CT poses a challenge for standardized radiomics analysis in clinical practice.

Purpose of the Study:

• To evaluate the influence of CT slice thickness on RF reproducibility and discriminative ability.
• To assess the efficacy of a deep learning-based CT slice synthesis (DLS) algorithm in overcoming limitations of thick-slice CT.

Main Methods:

• Retrospective analysis of 506 lung nodule CT scans (1-mm and 5-mm slice thickness) from two cohorts.
• Application of a DLS algorithm to synthesize 1-mm CT from 5-mm CT data.
• Extraction and comparison of RFs, reproducibility (CCC), and discriminative power (AUC) between real and synthesized CT images.

Main Results:

• DLS-synthesized 1-mm CT demonstrated significantly higher RF reproducibility compared to real 5-mm CT (p < 0.001).
• DLS 1-mm CT showed marked improvement in RF reproducibility (26.9%-27.6%) versus 5-mm CT (0.9%).
• The discriminative power of RFs from DLS 1-mm CT was superior to 5-mm CT and non-inferior to real 1-mm CT.

Conclusions:

• CT slice thickness critically affects radiomic feature reproducibility and discriminative performance.
• Deep learning-based CT slice synthesis effectively mitigates the negative impact of thick slices, supporting radiomics standardization and clinical translation.