3D Neighbor2Neighbor-based unsupervised deep learning for noise reduction in OCT imaging: insights from multiple...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Optical Coherence Tomography (OCT) is vital for non-invasive 3D imaging in ophthalmology and dermatology.
• Speckle and electrical noise degrade OCT image quality, hindering early diagnosis and treatment.
• Existing noise reduction methods may not fully leverage OCT's 3D volumetric data properties.

Purpose of the Study:

• To develop an unsupervised deep learning strategy for effective noise reduction in 3D OCT imaging.
• To optimize a ResNet network within a 3D framework to enhance noise reduction performance.
• To improve the clinical utility of OCT in ophthalmology and dermatology through enhanced image quality.

Main Methods:

• A three-dimensional (3D) Neighbor2Neighbor-based unsupervised deep learning framework was proposed.
• The ResNet network was optimized for improved noise reduction within the 3D OCT data.
• The method leverages inherent properties of 3D OCT volumetric data for noise suppression.

Main Results:

• The proposed method effectively reduced noise in 3D OCT images.
• Clearer tissue construct details were obtained, indicating enhanced image quality.
• Experimental results validated the noise reduction capabilities and potential clinical benefits.

Conclusions:

• The 3D deep learning strategy offers a powerful solution for noise reduction in OCT imaging.
• Optimized ResNet integration improves performance for clinical applications.
• This advancement supports broader and deeper applications of OCT in medical fields.