Related Experiment Video
Updated: Jan 6, 2026
06:03
Integrating Augmented Reality Tools in Breast Cancer Related Lymphedema Prognostication and Diagnosis
Published on: February 6, 2020
7.0K
Use of a Preliminary Artificial Intelligence-Based Laryngeal Cancer Screening Framework for Low-Resource Settings:
Shao Wei Sean Lam1, Min Hun Lee2, Michael Dorosan1
1Health Services Research Centre, Singapore Health Services Pte Ltd, Ngee Ann Kongsi Discovery Tower Level 6, 20 College Road, Singapore, 169856, Singapore, 65 65767140.
JMIR Formative Research
|October 7, 2025
Summary
An AI screening tool can help triage patients for laryngeal cancer in low-resource areas. This artificial intelligence framework efficiently identifies high-quality images from flexible nasopharyngoscopy videos for faster diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Healthcare
- Oncology
Background:
- Early diagnosis of laryngeal cancer is crucial for patient survival and quality of life.
- Limited access to specialists in low-resource settings impedes timely diagnosis using flexible nasopharyngoscopy (FNS) videos.
- Accurate triage of at-risk patients is hindered by the scarcity of expert interpretation of FNS videos.
Purpose of the Study:
- To introduce a preliminary AI-based screening framework for triaging at-risk patients in low-resource settings.
- To address challenges in analyzing high-dimensional FNS videos, including image selection, anatomical landmark identification, and patient classification.
- To develop a system capable of classifying patients into referral grades based on FNS video frames.
Main Methods:
- Developed an image quality model (IQM) to select high-quality endoscopic images from FNS videos.
- Utilized a disease classification model (DCM) employing efficient convolutional neural network (CNN) modules.
- Validated the approach using a real-world dataset of 132 patients from a US academic tertiary care center.
Main Results:
- The IQM achieved an AUROC of 0.895 and AUPRC of 0.878 for quality frame selection.
- The DCM improved performance by 38% (AUROC) and 8% (AUPRC) when using IQM-selected frames.
- An efficient CNN model demonstrated 2.5x faster inference time compared to ResNet50, with 50 frames needed for optimal results.
Conclusions:
- Demonstrated the feasibility of an AI-based screening framework for efficient patient triage in low-resource settings.
- The AI approach offers significant potential to improve healthcare accessibility and patient outcomes in underserved regions.
- This research provides foundational evidence for developing a fully validated screening system for laryngeal cancer.
