Optical coherence tomography (OCT) is widely used for diagnosing and monitoring retinal diseases, such as age-related macular degeneration (AMD). The segmentation of biomarkers such as layers and lesions is essential for patient diagnosis and follow-up. Recently, semi-supervised learning has shown promise in improving retinal segmentation performance. However, existing methods often produce anatomically implausible segmentations, fail to effectively model layer-lesion interactions, and lack guarantees on topological correctness. To address these limitations, we propose a novel semi-supervised model that introduces a fully differentiable biomarker topology engine to enforce anatomically correct segmentation of lesions and layers. This enables joint learning with bidirectional influence between layers and lesions, leveraging unlabeled and diverse partially labeled datasets. Our model learns a disentangled representation, separating spatial and style factors. This approach enables more realistic layer segmentations and improves lesion segmentation, while strictly enforcing lesion location in their anatomically plausible positions relative to the segmented layers. We evaluate the proposed model on public and internal datasets of OCT scans and show that it outperforms the current state-of-the-art in both lesion and layer segmentation, while demonstrating the ability to generalize layer segmentation to pathological cases using partially annotated training data. Our results demonstrate the potential of using anatomical constraints in semi-supervised learning for accurate, robust, and trustworthy retinal biomarker segmentation.

The rise of imaging techniques such as optical coherence tomography (OCT) and advances in deep learning (DL) have enabled clinicians and researchers to streamline the assessment of retinal diseases. A popular DL approach is self-supervised learning (SSL), where models learn from vast amounts of unlabeled data, avoiding costly annotation. SSL has allowed the development of foundation models (FMs), large models that can be used for a variety of downstream tasks. However, existing FMs for OCT, trained solely on image data, lack a comprehensive and robust semantic understanding of images, as evidenced by their downstream performance (especially for complex tasks), and thus require supervised fine-tuning (which may be unfeasible) to better adapt to specific applications and populations. To address this, we propose RetFiner, an SSL vision-language refinement scheme that improves the representations of existing FMs and enables their efficient and direct adaptation to specific populations for improved downstream performance. Our method uses a diverse set of training objectives which take advantage of the rich supervisory signal found in textual data. We tested RetFiner on the retinal FMs RETFound, UrFound, and VisionFM, showing significant improvements in linear probing performance on seven highly diverse OCT classification tasks, with an average increase of 5.7, 3.9, and 2.1% points over their baselines, respectively.

Deep learning has potential to automate screening, monitoring and grading of disease in medical images. Pretraining with contrastive learning enables models to extract robust and generalisable features from natural image datasets, facilitating label-efficient downstream image analysis. However, the direct application of conventional contrastive methods to medical datasets introduces two domain-specific issues. Firstly, several image transformations which have been shown to be crucial for effective contrastive learning do not translate from the natural image to the medical image domain. Secondly, the assumption made by conventional methods, that any two images are dissimilar, is systematically misleading in medical datasets depicting the same anatomy and disease. This is exacerbated in longitudinal image datasets that repeatedly image the same patient cohort to monitor their disease progression over time. In this paper we tackle these issues by extending conventional contrastive frameworks with a novel metadata-enhanced strategy. Our approach employs widely available patient metadata to approximate the true set of inter-image contrastive relationships. To this end we employ records for patient identity, eye position (i.e. left or right) and time series information. In experiments using two large longitudinal datasets containing 170,427 retinal optical coherence tomography (OCT) images of 7912 patients with age-related macular degeneration (AMD), we evaluate the utility of using metadata to incorporate the temporal dynamics of disease progression into pretraining. Our metadata-enhanced approach outperforms both standard contrastive methods and a retinal image foundation model in five out of six image-level downstream tasks related to AMD. We find benefits in both a low-data and high-data regime across tasks ranging from AMD stage and type classification to prediction of visual acuity. Due to its modularity, our method can be quickly and cost-effectively tested to establish the potential benefits of including available metadata in contrastive pretraining.

Automated lesion detection in retinal optical coherence tomography (OCT) scans has shown promise for several clinical applications, including diagnosis, monitoring and guidance of treatment decisions. However, segmentation models still struggle to achieve the desired results for some complex lesions or datasets that commonly occur in real-world, e.g. due to variability of lesion phenotypes, image quality or disease appearance. While several techniques have been proposed to improve them, one line of research that has not yet been investigated is the incorporation of additional semantic context through the application of anomaly detection models. In this study we experimentally show that incorporating weak anomaly labels to standard segmentation models consistently improves lesion segmentation results. This can be done relatively easy by detecting anomalies with a separate model and then adding these output masks as an extra class for training the segmentation model. This provides additional semantic context without requiring extra manual labels. We empirically validated this strategy using two in-house and two publicly available retinal OCT datasets for multiple lesion targets, demonstrating the potential of this generic anomaly guided segmentation approach to be used as an extra tool for improving lesion detection models.

The use of multimodal imaging has led to significant improvements in the diagnosis and treatment of many diseases. Similar to clinical practice, some works have demonstrated the benefits of multimodal fusion for automatic segmentation and classification using deep learning-based methods. However, current segmentation methods are limited to fusion of modalities with the same dimensionality (e.g., 3D + 3D, 2D + 2D), which is not always possible, and the fusion strategies implemented by classification methods are incompatible with localization tasks. In this work, we propose a novel deep learning-based framework for the fusion of multimodal data with heterogeneous dimensionality (e.g., 3D + 2D) that is compatible with localization tasks. The proposed framework extracts the features of the different modalities and projects them into the common feature subspace. The projected features are then fused and further processed to obtain the final prediction. The framework was validated on the following tasks: segmentation of geographic atrophy (GA), a late-stage manifestation of age-related macular degeneration, and segmentation of retinal blood vessels (RBV) in multimodal retinal imaging. Our results show that the proposed method outperforms the state-of-the-art monomodal methods on GA and RBV segmentation by up to 3.10% and 4.64% Dice, respectively.

The lack of reliable biomarkers makes predicting the conversion from intermediate to neovascular age-related macular degeneration (iAMD, nAMD) a challenging task. We develop a Deep Learning (DL) model to predict the future risk of conversion of an eye from iAMD to nAMD from its current OCT scan. Although eye clinics generate vast amounts of longitudinal OCT scans to monitor AMD progression, only a small subset can be manually labeled for supervised DL. To address this issue, we propose Morph-SSL, a novel Self-supervised Learning (SSL) method for longitudinal data. It uses pairs of unlabelled OCT scans from different visits and involves morphing the scan from the previous visit to the next. The Decoder predicts the transformation for morphing and ensures a smooth feature manifold that can generate intermediate scans between visits through linear interpolation. Next, the Morph-SSL trained features are input to a Classifier which is trained in a supervised manner to model the cumulative probability distribution of the time to conversion with a sigmoidal function. Morph-SSL was trained on unlabelled scans of 399 eyes (3570 visits). The Classifier was evaluated with a five-fold cross-validation on 2418 scans from 343 eyes with clinical labels of the conversion date. The Morph-SSL features achieved an AUC of 0.779 in predicting the conversion to nAMD within the next 6 months, outperforming the same network when trained end-to-end from scratch or pre-trained with popular SSL methods. Automated prediction of the future risk of nAMD onset can enable timely treatment and individualized AMD management.

Self-supervised learning (SSL) has emerged as a powerful technique for improving the efficiency and effectiveness of deep learning models. Contrastive methods are a prominent family of SSL that extract similar representations of two augmented views of an image while pushing away others in the representation space as negatives. However, the state-of-the-art contrastive methods require large batch sizes and augmentations designed for natural images that are impractical for 3D medical images. To address these limitations, we propose a new longitudinal SSL method, 3DTINC, based on non-contrastive learning. It is designed to learn perturbation-invariant features for 3D optical coherence tomography (OCT) volumes, using augmentations specifically designed for OCT. We introduce a new non-contrastive similarity loss term that learns temporal information implicitly from intra-patient scans acquired at different times. Our experiments show that this temporal information is crucial for predicting progression of retinal diseases, such as age-related macular degeneration (AMD). After pretraining with 3DTINC, we evaluated the learned representations and the prognostic models on two large-scale longitudinal datasets of retinal OCTs where we predict the conversion to wet-AMD within a six-month interval. Our results demonstrate that each component of our contributions is crucial for learning meaningful representations useful in predicting disease progression from longitudinal volumetric scans.

Robust forecasting of the future anatomical changes inflicted by an ongoing disease is an extremely challenging task that is out of grasp even for experienced healthcare professionals. Such a capability, however, is of great importance since it can improve patient management by providing information on the speed of disease progression already at the admission stage, or it can enrich the clinical trials with fast progressors and avoid the need for control arms by the means of digital twins. In this work, we develop a deep learning method that models the evolution of age-related disease by processing a single medical scan and providing a segmentation of the target anatomy at a requested future point in time. Our method represents a time-invariant physical process and solves a large-scale problem of modeling temporal pixel-level changes utilizing NeuralODEs. In addition, we demonstrate the approaches to incorporate the prior domain-specific constraints into our method and define temporal Dice loss for learning temporal objectives. To evaluate the applicability of our approach across different age-related diseases and imaging modalities, we developed and tested the proposed method on the datasets with 967 retinal OCT volumes of 100 patients with Geographic Atrophy and 2823 brain MRI volumes of 633 patients with Alzheimer's Disease. For Geographic Atrophy, the proposed method outperformed the related baseline models in the atrophy growth prediction. For Alzheimer's Disease, the proposed method demonstrated remarkable performance in predicting the brain ventricle changes induced by the disease, achieving the state-of-the-art result on TADPOLE cross-sectional prediction challenge dataset.

Bruch's membrane (BM) segmentation on optical coherence tomography (OCT) is a pivotal step for the diagnosis and follow-up of age-related macular degeneration (AMD), one of the leading causes of blindness in the developed world. Automated BM segmentation methods exist, but they usually do not account for the anatomical coherence of the results, neither provide feedback on the confidence of the prediction. These factors limit the applicability of these systems in real-world scenarios. With this in mind, we propose an end-to-end deep learning method for automated BM segmentation in AMD patients. An Attention U-Net is trained to output a probability density function of the BM position, while taking into account the natural curvature of the surface. Besides the surface position, the method also estimates an A-scan wise uncertainty measure of the segmentation output. Subsequently, the A-scans with high uncertainty are interpolated using thin plate splines (TPS). We tested our method with ablation studies on an internal dataset with 138 patients covering all three AMD stages, and achieved a mean absolute localization error of 4.10 μm. In addition, the proposed segmentation method was compared against the state-of-the-art methods and showed a superior performance on an external publicly available dataset from a different patient cohort and OCT device, demonstrating strong generalization ability.

Optical Coherence Tomography Angiography (OCTA), a functional extension of OCT, has the potential to replace most invasive fluorescein angiography (FA) exams in ophthalmology. So far, OCTA's field of view is however still lacking behind fluorescence fundus photography techniques. This is problematic, because many retinal diseases manifest at an early stage by changes of the peripheral retinal capillary network. It is therefore desirable to expand OCTA's field of view to match that of ultra-widefield fundus cameras. We present a custom developed clinical high-speed swept-source OCT (SS-OCT) system operating at an acquisition rate 8-16 times faster than today's state-of-the-art commercially available OCTA devices. Its speed allows us to capture ultra-wide fields of view of up to 90 degrees with an unprecedented sampling density and hence extraordinary resolution by merging two single shot scans with 60 degrees in diameter. To further enhance the visual appearance of the angiograms, we developed for the first time a three-dimensional deep learning based algorithm for denoising volumetric OCTA data sets. We showcase its imaging performance and clinical usability by presenting images of patients suffering from diabetic retinopathy.

The fovea centralis is an essential landmark in the retina where the photoreceptor layer is entirely composed of cones responsible for sharp, central vision. The localization of this anatomical landmark in optical coherence tomography (OCT) volumes is important for assessing visual function correlates and treatment guidance in macular disease. In this study, the "PRE U-net" is introduced as a novel approach for a fully automated fovea centralis detection, addressing the localization as a pixel-wise regression task. 2D B-scans are sampled from each image volume and are concatenated with spatial location information to train the deep network. A total of 5586 OCT volumes from 1,541 eyes were used to train, validate and test the deep learning method. The test data is comprised of healthy subjects and patients affected by neovascular age-related macular degeneration (nAMD), diabetic macula edema (DME) and macular edema from retinal vein occlusion (RVO), covering the three major retinal diseases responsible for blindness. Our experiments demonstrate that the PRE U-net significantly outperforms state-of-the-art methods and improves the robustness of automated localization, which is of value for clinical practice.

Neovascular age-related macular degeneration (nAMD) is nowadays successfully treated with anti-VEGF substances, but inter-individual treatment requirements are vastly heterogeneous and currently poorly plannable resulting in suboptimal treatment frequency. Optical coherence tomography (OCT) with its 3D high-resolution imaging serves as a companion diagnostic to anti-VEGF therapy. This creates a need for building predictive models using automated image analysis of OCT scans acquired during the treatment initiation phase. We propose such a model based on deep learning (DL) architecture, comprised of a densely connected neural network (DenseNet) and a recurrent neural network (RNN), trainable end-to-end. The method starts by sampling several 2D-images from an OCT volume to obtain a lower-dimensional OCT representation. At the core of the predictive model, the DenseNet learns useful retinal spatial features while the RNN integrates information from different time points. The introduced model was evaluated on the prediction of anti-VEGF treatment requirements in nAMD patients treated under a pro-re-nata (PRN) regimen. The DL model was trained on 281 patients and evaluated on a hold-out test set of 69 patient. The predictive model achieved a concordance index of 0.7 in regressing the number of received treatments, while in a classification task it obtained an 0.85 (0.81) AUC in detecting the patients with low (high) treatment requirements. The proposed model outperformed previous machine learning strategies that relied on a set of spatio-temporal image features, showing that the proposed DL architecture successfully learned to extract the relevant spatio-temporal patterns directly from raw longitudinal OCT images.

Diagnosis and treatment guidance are aided by detecting relevant biomarkers in medical images. Although supervised deep learning can perform accurate segmentation of pathological areas, it is limited by requiring a priori definitions of these regions, large-scale annotations, and a representative patient cohort in the training set. In contrast, anomaly detection is not limited to specific definitions of pathologies and allows for training on healthy samples without annotation. Anomalous regions can then serve as candidates for biomarker discovery. Knowledge about normal anatomical structure brings implicit information for detecting anomalies. We propose to take advantage of this property using Bayesian deep learning, based on the assumption that epistemic uncertainties will correlate with anatomical deviations from a normal training set. A Bayesian U-Net is trained on a well-defined healthy environment using weak labels of healthy anatomy produced by existing methods. At test time, we capture epistemic uncertainty estimates of our model using Monte Carlo dropout. A novel post-processing technique is then applied to exploit these estimates and transfer their layered appearance to smooth blob-shaped segmentations of the anomalies. We experimentally validated this approach in retinal optical coherence tomography (OCT) images, using weak labels of retinal layers. Our method achieved a Dice index of 0.789 in an independent anomaly test set of age-related macular degeneration (AMD) cases. The resulting segmentations allowed very high accuracy for separating healthy and diseased cases with late wet AMD, dry geographic atrophy (GA), diabetic macular edema (DME) and retinal vein occlusion (RVO). Finally, we qualitatively observed that our approach can also detect other deviations in normal scans such as cut edge artifacts.

Retinal swelling due to the accumulation of fluid is associated with the most vision-threatening retinal diseases. Optical coherence tomography (OCT) is the current standard of care in assessing the presence and quantity of retinal fluid and image-guided treatment management. Deep learning methods have made their impact across medical imaging, and many retinal OCT analysis methods have been proposed. However, it is currently not clear how successful they are in interpreting the retinal fluid on OCT, which is due to the lack of standardized benchmarks. To address this, we organized a challenge RETOUCH in conjunction with MICCAI 2017, with eight teams participating. The challenge consisted of two tasks: fluid detection and fluid segmentation. It featured for the first time: all three retinal fluid types, with annotated images provided by two clinical centers, which were acquired with the three most common OCT device vendors from patients with two different retinal diseases. The analysis revealed that in the detection task, the performance on the automated fluid detection was within the inter-grader variability. However, in the segmentation task, fusing the automated methods produced segmentations that were superior to all individual methods, indicating the need for further improvements in the segmentation performance.

The identification and quantification of markers in medical images is critical for diagnosis, prognosis, and disease management. Supervised machine learning enables the detection and exploitation of findings that are known a priori after annotation of training examples by experts. However, supervision does not scale well, due to the amount of necessary training examples, and the limitation of the marker vocabulary to known entities. In this proof-of-concept study, we propose unsupervised identification of anomalies as candidates for markers in retinal optical coherence tomography (OCT) imaging data without a constraint to a priori definitions. We identify and categorize marker candidates occurring frequently in the data and demonstrate that these markers show a predictive value in the task of detecting disease. A careful qualitative analysis of the identified data driven markers reveals how their quantifiable occurrence aligns with our current understanding of disease course, in early- and late age-related macular degeneration (AMD) patients. A multi-scale deep denoising autoencoder is trained on healthy images, and a one-class support vector machine identifies anomalies in new data. Clustering in the anomalies identifies stable categories. Using these markers to classify healthy-, early AMD- and late AMD cases yields an accuracy of 81.40%. In a second binary classification experiment on a publicly available data set (healthy versus intermediate AMD), the model achieves an area under the ROC curve of 0.944.

Prediction of treatment responses from available data is key to optimizing personalized treatment. Retinal diseases are treated over long periods and patients' response patterns differ substantially, ranging from a complete response to a recurrence of the disease and need for re-treatment at different intervals. Linking observable variables in high-dimensional observations to outcome is challenging. In this paper, we present and evaluate two different data-driven machine learning approaches operating in a high-dimensional feature space: sparse logistic regression and random forests-based extra trees (ET). Both identify spatio-temporal signatures based on retinal thickness features measured in longitudinal spectral-domain optical coherence tomography (OCT) imaging data and predict individual patient outcome using these quantitative characteristics. We demonstrate on a data set of monthly SD-OCT scans of 155 patients with central retinal vein occlusion (CRVO) and 92 patients with branch retinal vein occlusion (BRVO) followed over one year that we can predict from initial three observations if the treated disease will recur within the covered interval. ET predicts the outcome on fivefold cross-validation with an area under the receiver operating characteristic curve (AuC) of 0.83 for BRVO and 0.76 for CRVO. Logistic regression achieved an AuC of 0.78 and 0.79, respectively. At the same time, the methods identified stable predictive signatures in the longitudinal imaging data that are the basis for accurate prediction. Furthermore, our results show that taking spatio-temporal features into account improves accuracy compared with features extracted at a single time-point. Our results demonstrate the feasibility of mining longitudinal data for predictive signatures, and building predictive models based on observed data.