Photo Credit: Tamer Soliman
The following is a summary of “Prediction of functional and anatomical progression in lamellar macular holes,” published in the April 2024 issue of Ophthalmology by Crincoli et al.
AI can find markers in eye scans indicating the progress of lamellar macular holes (LMH). This information will be utilized to develop a Deep Learning (DL) model predicting vision loss.
Researchers conducted a retrospective study to create a DL model using Optical Coherence Tomography (OCT) and OCT angiography (OCTA) to predict visual acuity (VA) decline in untreated LMHs.
They recruited adult patients diagnosed with idiopathic LMHs with good-quality eye scans. The OCT and OCTA scans were used to train a DL model that can predict vision loss over two years. Biomarkers like the Ellipsoid zone (EZ), tissue loss(TL), and vitreopapillary adhesion (VPA) were analyzed. A regression model and SVM analysis were used with visualization maps to understand LMF progression.
The results showed that functionally progressing LMHs ( VA-PROG group, 41/139 eyes [29.5%]) had more damage, higher TL, more VPA, and lower superficial capillary plexus parafoveal vessel density (SCP VD) and vessel length density (VLD) compared to stable LMHs (VA-STABLE group, 98/139 eyes [70.5%]). DL and SVM models were 92.5% and 90.5% accurate. Top SVM features EZ damage, TL, CC FDD, and parafoveal SCP VD. Epiretinal proliferation and lower CMT were only linked to anatomical progression.
Investigators concluded that DL predicted LMH progression well over 2 years. While AI could enhance understanding retinal diseases’ natural path, CC and SCP integrity were crucial for LMH advancement.
Source: ophthalmologyscience.org/article/S2666-9145(24)00065-4/fulltext
