Systematic Review of Computer Vision Applications in Vascular Surgery

Overview

This systematic review analyzed 288 studies on computer vision technologies in vascular surgery, highlighting a rapid increase in research since 2017. Most studies focused on aortic pathologies, carotid stenosis, and foot ulcers, with limited attention to peripheral artery disease. Performance metrics varied, with dice coefficient and accuracy most common, but overall methodological quality and reporting adherence were suboptimal.

Background

Computer vision is an emerging field offering tools to enhance clinical practice in vascular surgery by enabling automated image analysis and disease detection. Vascular diseases such as aortic aneurysms, carotid stenosis, and diabetic foot ulcers present opportunities for computer vision applications. Despite growing research interest, the quality and consistency of studies vary, and standardized reporting guidelines like TRIPOD+AI are not widely followed. Understanding current trends and gaps is essential to guide future development and clinical integration.

Data Highlights

Key Findings

• There has been an exponential increase in computer vision studies in vascular surgery since 2017.
• The majority of research targets aortic pathologies (33%), carotid stenosis (30%), and diabetic foot ulcers (25%), with peripheral artery disease underrepresented (6%).
• Most studies are observational, predominantly retrospective (81%), with very few prospective designs and only one clinical trial.
• Dice coefficient (51%) and accuracy (36%) are the most frequently reported performance metrics; AUROC is less commonly used (17%).
• Only 15% of studies demonstrated a low risk of bias, and overall adherence to the TRIPOD+AI reporting guidelines was moderate at 57%.
• Recommendations include increased focus on peripheral artery disease, appropriate use of dice coefficient for segmentation tasks and AUROC for discrimination tasks, and early consultation of TRIPOD+AI guidelines during model development.

Clinical Implications

Clinicians should be aware that while computer vision technologies show promise in vascular surgery, current evidence is largely based on retrospective observational studies with variable quality. Greater emphasis on prospective validation and adherence to standardized reporting can improve reliability. Incorporating recommended performance metrics and focusing on underrepresented conditions like peripheral artery disease may enhance clinical applicability.

Conclusion

Computer vision applications in vascular surgery are rapidly expanding but require improved methodological rigor and reporting standards. Addressing current gaps will facilitate translation of these technologies into clinical practice.

References

1. Basu et al. 2020 -- Artificial intelligence: How is it changing medical sciences and its future?
2. Esteva et al. 2021 -- Deep learning-enabled medical computer vision
3. Javidan et al. 2022 -- A systematic review and bibliometric analysis of AI in vascular surgery
4. Li et al. 2022 -- Machine learning in vascular surgery: a systematic review and critical appraisal
5. Tomihama et al. 2023 -- Machine learning and image analysis in vascular surgery