Related Experiment Video
Updated: Jun 21, 2025
Author Spotlight: Bridging Gaps in Anatomy and Establishing a Foundation for Algorithmic Studies
Published on: December 15, 2023
Artificial intelligence for automatic detection and segmentation of nasal polyposis: a pilot study.
Vittorio Rampinelli1, Alberto Paderno2, Carlo Conti3
1Unit of Otorhinolaryngology - Head and Neck Surgery, Department of Surgical and Medical Specialties, Radiological Sciences, and Public Health, School of Medicine, ASST Spedali Civili, University of Brescia, Brescia, Italy. vittorio.rampinelli@gmail.com.
An AI algorithm effectively segments nasal polyps in Chronic rhinosinusitis with nasal polyposis (CRSwNP) patients using endoscopy videos. This tool aids in accurate diagnosis and treatment planning for CRSwNP.
Area of Science:
- Otorhinolaryngology
- Medical Imaging
- Artificial Intelligence
Background:
- Accurate diagnosis and quantification of nasal polyps are crucial for managing Chronic rhinosinusitis with nasal polyposis (CRSwNP).
- Current diagnostic methods may benefit from advanced image analysis tools.
Purpose of the Study:
- To develop an AI-based image analysis system for segmenting nasal polyps from nasal endoscopy videos.
- To evaluate the efficacy of an AI algorithm in identifying and delineating nasal polyps in CRSwNP patients.
Main Methods:
- Retrospective analysis of nasal videoendoscopies from 52 CRSwNP patients.
- Manual segmentation using Roboflow, followed by training of the Ultralytics YOLOv8.0.28 model on 342 images.
- Automated segmentation using the YOLOv8s-seg model.
Main Results:
- The YOLOv8s-seg model demonstrated high performance with a precision of 0.91 and mAP50 of 0.949 on the validation set.
- Segmentation task metrics showed a mean average precision (mAP) between 0.675 and 0.679 for IoUs from 50% to 95%.
- The AI model consists of 195 layers and requires 42.4 GFLOPs for operation.
Conclusions:
- A trained AI algorithm can effectively identify and delineate nasal polyps in CRSwNP patients.
- The AI system shows promise as a complementary tool to existing diagnostic methods for CRSwNP.
- Further research may address limitations such as the focus on CRSwNP-specific samples.
