Lesion Detection in Optical Coherence Tomography with Transformer-Enhanced Detector

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Deep Learning for Anomaly Detection

Background:

• Optical Coherence Tomography (OCT) is a valuable imaging technique used in ophthalmology and dentistry.
• Speckle noise in OCT images presents a significant challenge for accurate clinical diagnosis.
• Existing deep learning algorithms struggle with precise anomaly detection in OCT data.

Purpose of the Study:

• To propose a novel region-based, deep learning framework, Transformer-Enhanced Detection (TED), for anomaly detection in OCT images.
• To enhance diagnostic capabilities by accurately identifying and removing noise artifacts and other anomalies.
• To improve the clinical interpretation of OCT-acquired images across different medical specialties.

Main Methods:

• Developed Transformer-Enhanced Detection (TED), a deep learning framework incorporating attention gates (AGs).
• TED focuses on foreground identification while simultaneously detecting and removing noise artifacts as anomalies.
• Evaluated TED's performance on two dental OCT datasets and one CT dataset (lung nodules, livers).

Main Results:

• TED successfully detected tooth decay in dental datasets and various lesions in CT scans.
• Achieved superior performance compared to existing deep learning detection algorithms.
• Demonstrated significant improvements: 16-22% increase in accuracy and 10% in Intersection over Union (IOU) for dental datasets; 9% and 20% improvements for the CT dataset.

Conclusions:

• The proposed TED framework offers a robust solution for anomaly detection in OCT images.
• TED enhances diagnostic accuracy and aids clinical interpretation across multiple medical modalities.
• This deep learning approach shows significant potential for advancing medical image analysis.