Related Experiment Videos
An efficient framework for real-time colorectal polyp detection using local outlier factor-based preprocessing and
Saadat Behzadi1, Danial Sharifrazi2, Bita Mesbahzadeh3
1Department of Electrical, Electronic, and Information Engineering, University of Bologna, Bologna, Italy.
Scientific Reports
|July 16, 2026
Summary
This study introduces a fast and robust colorectal polyp detection framework using Local Outlier Factor (LOF) preprocessing and the YOLOv11n model. The method enhances training data quality, improving accuracy for early colorectal cancer diagnosis.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Colorectal cancer (CRC) prevention relies on timely and accurate polyp detection.
- Deep learning methods face challenges in automatic polyp detection due to image quality, polyp size, and data noise.
Purpose of the Study:
- To develop a lightweight, fast, and robust framework for colorectal polyp detection.
- To improve the quality of training data for object detection models.
Main Methods:
- A framework combining Local Outlier Factor (LOF) preprocessing with the YOLOv11n object detection model was proposed.
- Five public datasets (CVC-ColonDB, CVC-ClinicDB, Kvasir-SEG, ETIS, EndoScene) were utilized, converting segmentation masks to bounding boxes.
- LOF preprocessing was applied to the training set to remove noisy samples, followed by YOLOv11n training and five-fold cross-validation.
Main Results:
- The LOF-YOLOv11n framework achieved high performance: Precision (94.73%), Recall (91.46%), F1-score (93.28%), mAP@0.5 (96.54%), and mAP@0.5:0.95 (78.01%).
- The model demonstrated an average inference speed of 56.1 frames per second.
- LOF preprocessing showed limited improvements in learning stability and model performance without biasing evaluation.
Conclusions:
- Combining LOF-based training data improvement with the YOLOv11n detector offers a promising framework for computer-aided colorectal polyp detection.
- Further validation on independent clinical and video datasets is necessary before clinical deployment.