# Towards reliable calcification detection: calibration of uncertainty in coronary optical coherence tomography images

**Authors:** Hongshan Liu, Xueshen Li, Abdul Latif Bamba, Xiaoyu Song, Brigitta C. Brott, Silvio H. Litovsky, Yu Gan  
**arXiv:** [2211.06728](https://arxiv.org/abs/2211.06728)  
**Date:** 2022-11-12  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2211.06728v2/

## Key points

- Optical Coherence Tomography (OCT) is crucial in managing coronary artery disease (CAD)
- Percutaneous Coronary Intervention (PCI) relies on image-guided solutions for CAD treatment
- Accurate object detection is essential for identifying calcifications in arteries before treatment
- Deep learning-based object detection methods, such as You-Only-Look-Once v5 (YOLO), are used for detecting calcifications in coronary OCT images
- Calibrated confidence scores improve reliability of predictions and enhance clinical evaluations during CAD treatment

## Summary

, , , , 

In the field of medical imaging, Optical Coherence Tomography (OCT) has emerged as a crucial tool in the management of coronary artery disease (CAD). Percutaneous Coronary Intervention (PCI), a common procedure used in CAD treatment, relies heavily on image-guided solutions to navigate through narrowed arteries. However, the presence of unidentified calcified regions within these arteries can pose challenges and potentially impact the success of PCI outcomes. To address this issue, accurate object detection is essential for identifying the location and thickness of calcifications prior to treatment. Deep learning-based object detection methods have been increasingly explored across various applications, including medical imaging. The quality of object detection predictions plays a critical role in ensuring reliable results, particularly in safety-critical scenarios such as CAD treatment. In this study, researchers implemented the You-Only-Look-Once v5 (YOLO) object detection model within a framework designed specifically for detecting calcifications in coronary OCT images. One key aspect addressed in this research is the evaluation of uncertainty in prediction results. By assessing expected calibration errors, researchers were able to determine the certainty level of detection outcomes. To enhance the reliability of predictions, confidence scores were calibrated using dependent logistic calibration based on each detection result's confidence and center coordinates. This approach aimed to reduce uncertainty in calcification detection and improve the overall accuracy of results. The findings revealed that YOLO demonstrated higher precision and recall compared to other object detection models while producing more reliable outcomes. Through calibrated confidence scores, researchers achieved a confidence error rate of approximately 0.13, indicating improved trustworthiness in prediction results. This calibration process holds significant potential for enhancing clinical evaluations during CAD treatment procedures guided by imaging techniques. Authors involved in this study include Hongshan Liu, Xueshen Li, Abdul Latif Bamba, Xiaoyu Song, Brigitta C. Brott, Silvio H. Litovsky, and Yu Gan. The research paper is published under DOI 10.1117/1.JBO.28.3.036008 and provides valuable insights into refining calcification detection methods within coronary OCT imaging for improved clinical decision-making in CAD management scenarios.