Related Experiment Video
Updated: Jun 25, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.8K
Deep Learning and High-Resolution Anoscopy: Development of an Interoperable Algorithm for the Detection and
Miguel Mascarenhas Saraiva1,2,3, Lucas Spindler4, Thiago Manzione5
1Department of Gastroenterology, São João University Hospital, Alameda Professor Hernâni Monteiro, 4200-427 Porto, Portugal.
Cancers
|May 25, 2024
Summary
Artificial intelligence (AI) using deep learning accurately detects high-grade squamous intraepithelial lesions (HSIL) versus low-grade squamous intraepithelial lesions (LSIL) in high-resolution anoscopy (HRA) images. This AI system enhances early diagnosis of anal cancer precursors.
Area of Science:
- Gastroenterology
- Oncology
- Medical Imaging
Background:
- High-resolution anoscopy (HRA) is crucial for detecting anal squamous cell carcinoma (ASCC) precursors.
- Artificial intelligence (AI) shows promise in analyzing HRA images for lesion differentiation.
- Existing methods require expert interpretation, highlighting the need for automated solutions.
Purpose of the Study:
- To develop and evaluate a deep learning system for automatic detection and differentiation of HSIL versus LSIL using HRA images.
- To assess the AI system's performance across conventional and digital HRA platforms.
- To determine the AI's accuracy in various staining and post-treatment conditions.
Main Methods:
- A convolutional neural network (CNN) was developed using 57,822 HRA images from 151 exams.
- The dataset comprised images from conventional and digital proctoscopes.
- Performance was evaluated on subsets with acetic acid and Lugol iodine staining, and after therapeutic manipulation.
Main Results:
- The CNN achieved an overall accuracy of 94.6% in distinguishing HSIL from LSIL.
- The algorithm demonstrated high sensitivity (93.6%) and specificity (95.7%) with an AUC of 0.97.
- Accuracy remained high across different conditions, reaching 99.3% after therapeutic manipulation.
Conclusions:
- AI algorithms, specifically this CNN, can effectively enhance the early diagnosis of ASCC precursors.
- The developed AI system performs adequately on both conventional and digital HRA interfaces.
- Automated analysis of HRA images holds significant potential for improving ASCC precursor detection and management.
