Are Macula or Optic Nerve Head Structures Better at Diagnosing Glaucoma? An Answer Using Artificial Intelligence and...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma diagnosis relies on detecting structural changes in the optic nerve head (ONH) and macula.
• Current optical coherence tomography (OCT) imaging may not capture all relevant diagnostic information.
• Advancements in deep learning offer potential for automated analysis of OCT scans.

Purpose of the Study:

• To develop a deep-learning algorithm for automatic segmentation of ONH and macula structures in 3D wide-field OCT scans.
• To evaluate the diagnostic power of 3D ONH, macula, and combined structures for glaucoma detection.

Main Methods:

• A deep-learning algorithm was developed and trained on manually annotated OCT B-scans for structure segmentation.
• A 3D convolutional neural network (3D-CNN) was designed for glaucoma classification using segmented OCT volumes.
• The classification algorithm was tested on macular, ONH, and wide-field OCT scan datasets, with performance measured by AUC.

Main Results:

• The segmentation algorithm achieved a high Dice coefficient (DC) of 0.94 ± 0.003.
• The 3D-CNN demonstrated superior glaucoma classification performance using wide-field OCT scans (AUC = 0.99 ± 0.01).
• ONH and macula scans showed lower AUCs (0.93 ± 0.06 and 0.91 ± 0.11, respectively).

Conclusions:

• Wide-field OCT scans significantly improve glaucoma diagnosis compared to conventional ONH or macula OCTs.
• Automated deep learning segmentation and classification enhance diagnostic accuracy.
• This technology holds promise for widespread clinical adoption in glaucoma management.