Advancements in biomedical rendering: A survey on AI-based denoising techniques

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Deep learning-based denoising for Monte Carlo (MC) Path Tracing in computed tomography (CT) volume visualization shows promise but has inconsistent qualitative results.
• Evaluating the qualitative impact of AI on CT image denoising requires understanding user perception beyond standard metrics.

Purpose of the Study:

• To investigate the reasons for incongruent quantitative and qualitative assessments of deep learning-based denoising in CT volume visualization.
• To assess healthcare professionals' perceptions of AI-enhanced CT image and video quality, confidence, and clinical applicability.

Main Methods:

• A web-based SurveyMonkey questionnaire was distributed to radiologists, residents, orthopedic surgeons, and veterinarians.
• Participants evaluated randomized sections on AI-enhanced image/video quality, confidence in reference images, and clinical applicability.
• Seventy-four participants with varying experience levels (<1 to >5 years) completed the survey.

Main Results:

• A majority (77%) preferred AI-enhanced images over traditional MC estimates, with preference influenced by experience (adjusted OR 0.81).
• Participant experience correlated positively with confidence in AI-generated images and satisfaction with video previews (adjusted OR 0.96-0.98).
• Significant monotonic relationships were found between experience, confidence, and satisfaction.

Conclusions:

• AI post-processing holds potential for improving biomedical volume rendering in CT, particularly for experienced users.
• User preferences for AI-enhanced CT images may not always align with objective quality metrics like PSNR and SSIM.
• The study highlights the importance of qualitative assessments and user experience in evaluating AI's impact on medical image denoising.