Semivariogram and Semimadogram functions as descriptors for AMD diagnosis on SD-OCT topographic maps using Support...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Age-related macular degeneration (AMD) is a leading cause of blindness due to drusen formation in the macula.
• Current automated detection methods using spectral domain optical coherence tomography (SD-OCT) often analyze slices independently, losing spatial correlation.
• Retinal topography shares similarities with geographic maps, suggesting geostatistical analysis as a potential approach.

Purpose of the Study:

• To develop and evaluate a novel methodology for the automatic diagnosis of AMD using geostatistical functions on SD-OCT images.
• To overcome the limitations of slice-by-slice analysis in current automated AMD detection techniques.

Main Methods:

• Constructing topographic maps of the macular region from SD-OCT volumes.
• Computing geostatistical features (semivariogram, semimadogram) as texture descriptors from these maps.
• Utilizing a Support Vector Machine classifier trained with these geostatistical features.

Main Results:

• A database of 384 SD-OCT exams (269 AMD, 115 control) was used for training and testing.
• The best classification model achieved 95.2% accuracy and an AUROC of 0.989.
• The methodology demonstrated high performance in distinguishing between eyes with and without AMD.

Conclusions:

• Geostatistical descriptors provide an effective means for automated AMD diagnosis from SD-OCT data.
• The proposed method offers a competitive and promising alternative to existing techniques.
• This approach leverages the spatial information within SD-OCT volumes for improved diagnostic accuracy.