Abstract
Purpose :
The inherent local and global characteristics of visual fields (VFs) can be exploited in a strong data-driven sense and could provide better understanding of VFs with regards to glaucoma. Ultimately, this may help to efficiently automatize the diagnosis process. Our hypothesis is that alternative representations of raw VFs, in terms of different spatial scales, could be learned by computers using machine learning techniques towards an effective automatized glaucoma identification task. Accordingly, we present a Convolutional Neural Network (CNN)-based approach for classification of VFs as being glaucomatous or non-glaucomatous.
Methods :
We used two datasets. One dataset (D1) consists of VFs measured with Octopus (Haag-Streit AG, Koeniz, Switzerland) containing 44538 VFs from a mixed population of 14849 patients (i.e. healthy patients as well as patients of different and unknown visual impairments). The diagnosis for VFs in this dataset was determined using a custom implementation of the Glaucoma Hemifield Test. Accordingly 40738 VFs were found to be glaucomatous whereas 3800 VFs were normal. The second dataset (D2) includes 4863 VFs from 139 glaucomatous subjects and 245 VFs from 22 healthy subjects, acquired with Humphrey Visual Field Analyzer (Carl Zeiss Meditec AG, Jena, Germany). In our method, we first created voronoi diagrams out of the VFs in both datasets, which produce image representations of the VFs. Then, we constructed a CNN with 3 convolutional layers, 1 max-pooling layer and 3 fully connected layers and trained it on D1. The trained CNN was tested on D2 and its classification performance was measured using the False Positive Rates (FPR), area under Receiver Operating Characteristic (ROC) and Precision-Recall (RC) curves.
Results :
We obtain 0.932 ±0.0028 area under ROC curve, 0.996 ±0.0002 area under PR curve for the CNN classifier tested on dataset D2. Moreover, for a FNR of 0.05, the FPR was found to be 0.417 ±0.02.
Conclusions :
These results support the fact that processing VFs through a CNN generates different representation of data in terms of its hidden characteristics and patterns that are efficient to discriminate between glaucomatous and non-glaucomatous VFs in an automated way. The performance could be further improved with a different CNN architecture. The trained CNNs have the potential to be utilized for glaucoma progression analysis as well since they are successful in providing better representation of VFs.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.