Modeling Disease Progression in Retinal OCTs with Longitudinal Self-supervised Learning

Abstract

Longitudinal imaging is capable of capturing the static anatomical structures and the dynamic changes of the morphology resulting from aging or disease progression. Self-supervised learning allows to learn new representation from available large unlabelled data without any expert knowledge. We propose a deep learning self-supervised approach to model disease progression from longitudinal retinal optical coherence tomography (OCT). Our self-supervised model takes benefit from a generic time-related task, by learning to estimate the time interval between pairs of scans acquired from the same patient. This task is (i) easy to implement, (ii) allows to use irregularly sampled data, (iii) is tolerant to poor registration, and (iv) does not rely on additional annotations. This novel method learns a representation that focuses on progression specific information only, which can be transferred to other types of longitudinal problems. We transfer the learnt representation to a clinically highly relevant task of predicting the onset of an advanced stage of age-related macular degeneration within a given time interval based on a single OCT scan. The boost in prediction accuracy, in comparison to a network learned from scratch or transferred from traditional tasks, demonstrates that our pretrained self-supervised representation learns a clinically meaningful information.

Role of Deep Learning–Quantified Hyperreflective Foci for the Prediction of Geographic Atrophy Progression

“…Apart from robust retinal layer segmentation and the precise spatial differentiation and quantification of retinal fluid compartments, [50][51][52] more refined biomarkers on a subclinical level, such as ellipsoid zone thickness or HRF, and the interaction of these biomarkers can now be explored to assess individual risk in disease progression. 16,[53][54][55][56] The possibilities in AI are continuously expanding and novel, clinically unknown biomarkers can be identified using unsupervised learning. The latter lets the AI algorithm detect retinal markers on its own and evaluates them for the prediction of disease progression, which widely opens the spectrum of pathologic feature detection.…”

Imaging and artificial intelligence for progression of age-related macular degeneration

“…The inclusion criteria for studies utilizing learning algorithms for AMD progression were as follows: (i) studies published in the last decade, (ii) use of machine and/or deep learning methods for predictive modeling, (iii) an end point of progression to neovascular AMD or advanced dry AMD, and (iv) the inclusion of imaging data/imaging biomarkers. Of the 15 studies included, 7 used color fundus photos, 12,15,34,36,37,41,43 7 used SD-OCT scans, [23][24][25][38][39][40]42 and 1 used both image modalities 35 as input to ML and DL systems for AMD progression prediction (Table 2). Wu et al, 35 Yim et al, 38 and Burlina et al 41 used imaging only, while others included features representing demographic, environmental, genetics, clinical and temporal characteristics of patients.…”

“…15,23,34,37,43 Of those studies that did report AUC metrics, performance ranged from 0.68 to 0.97 (Table 2). We highlighted six studies that looked at progression to neovascular AMD, 15,[23][24][25]38,40 one that looked at progression to GA only, 39 and eight that included both. 12,[34][35][36][37][41][42][43] Progression predictions at different timepoints were reported, and were as early as three months and up to 12 years (Table 2).…”

“…12,[34][35][36][37][41][42][43] Progression predictions at different timepoints were reported, and were as early as three months and up to 12 years (Table 2). For studies that had one-to two-year follow-up, performance metrics were reported at threemonth intervals, 23 six-month intervals, 39 or yearly. 34 For studies with more than five-year follow-up, performance Table 1.…”

“…While the liberal selection of higher Sn provided a Sp of 0.55, the conservative selection of higher Sp provided a poor Sn of 0.34. 38 Rivail et al 39 predicted GA conversion in 6, 12, and 18 months with their deep Siamese network, a self-supervised learning system of spatiotemporal representations with 6-fold cross validation. AUC and precision were reported with the former remaining consistent across time and the latter increasing across time points 39 (Table 2).…”