Prediction of treatment outcome in neovascular age-related macular degeneration using a novel convolutional neural network

Abstract

While prognosis and risk of progression are crucial in developing precise therapeutic strategy in neovascular age-related macular degeneration (nAMD), limited predictive tools are available. We proposed a novel deep convolutional neural network that enables feature extraction through image and non-image data integration to seize imperative information and achieve highly accurate outcome prediction. The Heterogeneous Data Fusion Net (HDF-Net) was designed to predict visual acuity (VA) outcome (improvement ≥ 2 line or not) at 12th months after anti-VEGF treatment. A set of pre-treatment optical coherence tomography (OCT) image and non-image demographic features were employed as input data and the corresponding 12th-month post-treatment VA as the target data to train, validate, and test the HDF-Net. This newly designed HDF-Net demonstrated an AUC of 0.989 (95% CI 0.970–0.999), accuracy of 0.936 [95% confidence interval (CI) 0.889–0.964], sensitivity of 0.933 (95% CI 0.841–0.974), and specificity of 0.938 (95% CI 0.877–0.969). By simulating the clinical decision process with mixed pre-treatment information from raw OCT images and numeric data, HDF-Net demonstrated promising performance in predicting individualized treatment outcome. The results highlight the potential of deep learning to simultaneously process a broad range of clinical data to weigh and leverage the complete information of the patient. This novel approach is an important step toward real-world personalized therapeutic strategy for typical nAMD.

An overview of artificial intelligence in diabetic retinopathy and other ocular diseases

“…Detection of ARMD and its biomarkers from OCTA images via CNNs has tremendous potential to expedite the screening of patients with early and late-stage ARMD. Yeh et al ( 76 ) proposed a heterogeneous data fusion network (HDF-Net) to predict visual acuity (VA) and to evaluate the prognosis and risk of progression of neovascular age-related macular degeneration (nARMD). The clinical decision-making process was simulated using a mixture of pre-processed information from raw OCT images and digital data, and HDF-Net performed well in predicting individualized treatment outcomes.…”

Rao, Anil; Chandra, Shruti; Sivaprasad, Sobha; (2022) Prediction of treatment response in patients with neovascular age-related macular degeneration

“…An increase in VA of 10 letters was chosen as the cutpoint because it is considered to represent a clinically meaningful improvement in subjects with advanced eye disease [12]. This corresponds to an increase in VA of two lines of the Snellen chart, which is the cutpoint used in [23] for prediction of treatment response from OCT images and clinical/demographic variables.…”

“…The CNN architecture shown in figure 2 differs to the typical Deep Learning architectures used in Ophthalmology for prediction. Usually, 2D CNNs are used in which the model input is a single b-scan from a subject [3,23,16,17]. Such approaches require a way of pooling the model predictions from each b-scan to give a volume-level prediction.…”

“…Deep Learning has been used in a number of prediction tasks involving OCT, such as disease classification [14,3], prediction of conversion to nAMD [24], and treatment referral [16,11]. In [23], Deep Learning was used to predict treatment response of patients with nAMD from OCT and clinical/demographic variables. Although the trained models achieved high predictive accuracy, the study design differed from that of this work, and the described approach used a patented Neural Network architecture which therefore has a restricted use.…”

“…In the second approach, fully trained Deep Learning models are developed using only the OCT images. In contrast with [23], we use outcome measures based on visual acuity/changes in visual acuity 4-10 weeks after the last loading phase injection. The models are developed on a random training sample of data from a multi-site study, and evaluated on the remaining unseen test sample.…”