Machine learning in foveal avascular zone area measurement in patients undergoing cataract surgery
Author: Maria Soledad Pighin Caminos (Spain)
Co-authors: Evangelos Tsiroukis, Ignasi Jürgens
Purpose
To compare different measurement methods of the foveal avascular zone in OCT-A.
Setting/Venue
Institut Català de Retina
Methods
Seventy hundred and twenty images of sixty patients were evaluated. An OCT-A was performed in patients undergoing cataract surgery before and after treatment (one week, one month and two months). The deep, intermediate, and superficial plexuses were separately measured using the 3x3 mm macular cube OCT-A protocol of the Spectralis device (Heidelberg, Germany). Projection artifacts were removed, and images were measured with ImageJ using a semi-automated protocol and a machine learning software designed for ImageJ (Weka).
Results
The FAZ area of the superficial vascular plexus measured 0.44 mm2 when using the semi-automated protocol and 0.4 mm2 with the Weka protocol. Differences before and after cataract surgery were found with both protocols. The FAZ area of the intermediate vascular plexus was 0.16 mm2 in the semi-automated protocol and 0.19 mm2 in the Weka protocol. No differences were observed during follow-up visits using both protocols. The FAZ area of the deep vascular plexus was 0.38 mm2 in the semi-automated protocol and 0.44 mm2 in the Weka protocol. Differences before and after cataract surgery were found when using the Weka protocol.
Conlusions
Automated and semi-automated measurements of the FAZ area show more consistent results than manual measurements. Among automated protocols, machine learning based programs are gaining field of action. The Weka trainable segmentation program is useful and reliable in the FAZ measurement and it is not inferior to other protocols.