RESUMEN

Deep learning-based methods for fast target segmentation of computed tomography (CT) imaging have become increasingly popular. The success of current deep learning methods usually depends on a large amount of labeled data. Labeling medical data is a time-consuming and laborious task. Therefore, this paper aims to enhance the segmentation of CT images by using a semi-supervised learning method. In order to utilize the valid information in unlabeled data, we design a semi-supervised network model for contrastive learning based on entropy constraints. We use CNN and Transformer to capture the image's local and global feature information, respectively. In addition, the pseudo-labels generated by the teacher networks are unreliable and will lead to degradation of the model performance if they are directly added to the training. Therefore, unreliable samples with high entropy values are discarded to avoid the model extracting the wrong features. In the student network, we also introduce the residual squeeze and excitation module to learn the connection between different channels of each layer feature to obtain better segmentation performance. We demonstrate the effectiveness of the proposed method on the COVID-19 CT public dataset. We mainly considered three evaluation metrics: DSC, HD95, and JC. Compared with several existing state-of-the-art semi-supervised methods, our method improves DSC by 2.3%, JC by 2.5%, and reduces HD95 by 1.9 mm. In this paper, a semi-supervised medical image segmentation method is designed by fusing CNN and Transformer and utilizing entropy-constrained contrastive learning loss, which improves the utilization of unlabeled medical images.

RESUMEN

Fetal electrocardiogram (FECG) contains crucial information about the fetus during pregnancy, making the extraction of FECG signal essential for monitoring fetal health. However, extracting FECG signal from abdominal electrocardiogram (AECG) poses several challenges: (1) FECG signal is often contaminated by noise, and (2) FECG signal is frequently overshadowed by high-amplitude maternal electrocardiogram (MECG). To address these issues and enhance the accuracy of signal extraction, this paper proposes an improved Cycle Generative Adversarial Networks (CycleGAN) with integrated contrastive learning for FECG signal extraction. The model introduces a dual-attention mechanism in the generator of the generative adversarial network, incorporating a multi-head self-attention (MSA) module and a channel-wise self-attention (CSA) module to enhance the quality of generated signals. Additionally, a contrastive triplet loss is integrated into the CycleGAN loss function, optimizing training to increase the similarity between the extracted FECG signal and the scalp fetal electrocardiogram. The proposed method is evaluated using the ADFECG dataset and the PCDB dataset both from the Physionet. In terms of signal extraction quality, Mean Squared Error is reduced to 0.036, Mean Absolute Error (MAE) to 0.009, and Pearson Correlation Coefficient reaches 0.924. When validating the model performance, Structural Similarity Index achieves 95.54%, Peak Signal-to-Noise Ratio (PSNR) reaches 38.87 dB, and R-squared (R2) attains 95.12%. Furthermore, the positive predictive value (PPV), sensitivity (SEN) and F1-score for QRS wave cluster detection on the ADFECG dataset also reached 99.56%, 99.43% and 99.50%, respectively. On the PCDB dataset, the positive predictive value (PPV), sensitivity (SEN) and F1-score for QRS wave cluster detection also reached 98.24%, 98.60% and 98.42%, respectively. All of them are higher than other methods. Therefore, the proposed model has important applications in effective monitoring of fetal health during pregnancy.

RESUMEN

Despite significant advances in the deep clustering research, there remain three critical limitations to most of the existing approaches. First, they often derive the clustering result by associating some distribution-based loss to specific network layers, neglecting the potential benefits of leveraging the contrastive sample-wise relationships. Second, they frequently focus on representation learning at the full-image scale, overlooking the discriminative information latent in partial image regions. Third, although some prior studies perform the learning process at multiple levels, they mostly lack the ability to exploit the interaction between different learning levels. To overcome these limitations, this paper presents a novel deep image clustering approach via Partial Information discrimination and Cross-level Interaction (PICI). Specifically, we utilize a Transformer encoder as the backbone, coupled with two types of augmentations to formulate two parallel views. The augmented samples, integrated with masked patches, are processed through the Transformer encoder to produce the class tokens. Subsequently, three partial information learning modules are jointly enforced, namely, the partial information self-discrimination (PISD) module for masked image reconstruction, the partial information contrastive discrimination (PICD) module for the simultaneous instance- and cluster-level contrastive learning, and the cross-level interaction (CLI) module to ensure the consistency across different learning levels. Through this unified formulation, our PICI approach for the first time, to our knowledge, bridges the gap between the masked image modeling and the deep contrastive clustering, offering a novel pathway for enhanced representation learning and clustering. Experimental results across six image datasets demonstrate the superiority of our PICI approach over the state-of-the-art. In particular, our approach achieves an ACC of 0.772 (0.634) on the RSOD (UC-Merced) dataset, which shows an improvement of 29.7% (24.8%) over the best baseline. The source code is available at https://github.com/Regan-Zhang/PICI.

RESUMEN

A vast amount of single-cell RNA sequencing (SC) data have been accumulated via various studies and consortiums, but the lack of spatial information limits its analysis of complex biological activities. To bridge this gap, we introduce CellContrast, a computational method for reconstructing spatial relationships among SC cells from spatial transcriptomics (ST) reference. By adopting a contrastive learning framework and training with ST data, CellContrast projects gene expressions into a hidden space where proximate cells share similar representation values. We performed extensive benchmarking on diverse platforms, including SeqFISH, Stereo-seq, 10X Visium, and MERSCOPE, on mouse embryo and human breast cells. The results reveal that CellContrast substantially outperforms other related methods, facilitating accurate spatial reconstruction of SC. We further demonstrate CellContrast's utility by applying it to cell-type co-localization and cell-cell communication analysis with real-world SC samples, proving the recovered cell locations empower more discoveries and mitigate potential false positives.

RESUMEN

Aiming at the problem that the feature extraction ability of forehead single-channel electroencephalography (EEG) signals is insufficient, which leads to decreased fatigue detection accuracy, a fatigue feature extraction and classification algorithm based on supervised contrastive learning is proposed. Firstly, the raw signals are filtered by empirical modal decomposition to improve the signal-to-noise ratio. Secondly, considering the limitation of the one-dimensional signal in information expression, overlapping sampling is used to transform the signal into a two-dimensional structure, and simultaneously express the short-term and long-term changes of the signal. The feature extraction network is constructed by depthwise separable convolution to accelerate model operation. Finally, the model is globally optimized by combining the supervised contrastive loss and the mean square error loss. Experiments show that the average accuracy of the algorithm for classifying three fatigue states can reach 75.80%, which is greatly improved compared with other advanced algorithms, and the accuracy and feasibility of fatigue detection by single-channel EEG signals are significantly improved. The results provide strong support for the application of single-channel EEG signals, and also provide a new idea for fatigue detection research.

Asunto(s)

RESUMEN

Semi-supervised learning (SSL) has achieved significant success due to its capacity to alleviate annotation dependencies. Most existing SSL methods utilize pseudo-labeling to propagate useful supervised information for training unlabeled data. However, these methods ignore learning temporal representations, making it challenging to obtain a well-separable feature space for modeling explicit class boundaries. In this work, we propose a semi-supervised Time Series classification framework via Bidirectional Consistency with Temporal-aware (TS-BCT), which regularizes the feature space distribution by learning temporal representations through pseudo-label-guided contrastive learning. Specifically, TS-BCT utilizes time-specific augmentation to transform the entire raw time series into two distinct views, avoiding sampling bias. The pseudo-labels for each view, generated through confidence estimation in the feature space, are then employed to propagate class-related information into unlabeled samples. Subsequently, we introduce a temporal-aware contrastive learning module that learns discriminative temporal-invariant representations. Finally, we design a bidirectional consistency strategy by incorporating pseudo-labels from two distinct views into temporal-aware contrastive learning to construct a class-related contrastive pattern. This strategy enables the model to learn well-separated feature spaces, making class boundaries more discriminative. Extensive experimental results on real-world datasets demonstrate the effectiveness of TS-BCT compared to baselines.

RESUMEN

The integration of morphological attributes extracted from histopathological images and genomic data holds significant importance in advancing tumor diagnosis, prognosis, and grading. Histopathological images are acquired through microscopic examination of tissue slices, providing valuable insights into cellular structures and pathological features. On the other hand, genomic data provides information about tumor gene expression and functionality. The fusion of these two distinct data types is crucial for gaining a more comprehensive understanding of tumor characteristics and progression. In the past, many studies relied on single-modal approaches for tumor diagnosis. However, these approaches had limitations as they were unable to fully harness the information from multiple data sources. To address these limitations, researchers have turned to multi-modal methods that concurrently leverage both histopathological images and genomic data. These methods better capture the multifaceted nature of tumors and enhance diagnostic accuracy. Nonetheless, existing multi-modal methods have, to some extent, oversimplified the extraction processes for both modalities and the fusion process. In this study, we presented a dual-branch neural network, namely SG-Fusion. Specifically, for the histopathological modality, we utilize the Swin-Transformer structure to capture both local and global features and incorporate contrastive learning to encourage the model to discern commonalities and differences in the representation space. For the genomic modality, we developed a graph convolutional network based on gene functional and expression level similarities. Additionally, our model integrates a cross-attention module to enhance information interaction and employs divergence-based regularization to enhance the model's generalization performance. Validation conducted on glioma datasets from the Cancer Genome Atlas unequivocally demonstrates that our SG-Fusion model outperforms both single-modal methods and existing multi-modal approaches in both survival analysis and tumor grading.

RESUMEN

Unraveling the intricate network of associations among microRNAs (miRNAs), genes, and diseases is pivotal for deciphering molecular mechanisms, refining disease diagnosis, and crafting targeted therapies. Computational strategies, leveraging link prediction within biological graphs, present a cost-efficient alternative to high-cost empirical assays. However, while plenty of methods excel at predicting specific associations, such as miRNA-disease associations (MDAs), miRNA-target interactions (MTIs), and disease-gene associations (DGAs), a holistic approach harnessing diverse data sources for multifaceted association prediction remains largely unexplored. The limited availability of high-quality data, as vitro experiments to comprehensively confirm associations are often expensive and time-consuming, results in a sparse and noisy heterogeneous graph, hindering an accurate prediction of these complex associations. To address this challenge, we propose a novel framework called Global-local aware Heterogeneous Graph Contrastive Learning (GlaHGCL). GlaHGCL combines global and local contrastive learning to improve node embeddings in the heterogeneous graph. In particular, global contrastive learning enhances the robustness of node embeddings against noise by aligning global representations of the original graph and its augmented counterpart. Local contrastive learning enforces representation consistency between functionally similar or connected nodes across diverse data sources, effectively leveraging data heterogeneity and mitigating the issue of data scarcity. The refined node representations are applied to downstream tasks, such as MDA, MTI, and DGA prediction. Experiments show GlaHGCL outperforming state-of-the-art methods, and case studies further demonstrate its ability to accurately uncover new associations among miRNAs, genes, and diseases. We have made the datasets and source code publicly available at https://github.com/Sue-syx/GlaHGCL.

Asunto(s)

RESUMEN

Multi-view multi-label learning (MVML) aims to train a model that can explore the multi-view information of the input sample to obtain its accurate predictions of multiple labels. Unfortunately, a majority of existing MVML methods are based on the assumption of data completeness, making them useless in practical applications with partially missing views or some uncertain labels. Recently, many approaches have been proposed for incomplete data, but few of them can handle the case of both missing views and labels. Moreover, these few existing works commonly ignore potentially valuable information about unknown labels or do not sufficiently explore latent label information. Therefore, in this paper, we propose a label semantic-guided contrastive learning method named LSGC for the dual incomplete multi-view multi-label classification problem. Concretely, LSGC employs deep neural networks to extract high-level features of samples. Inspired by the observation of exploiting label correlations to improve the feature discriminability, we introduce a graph convolutional network to effectively capture label semantics. Furthermore, we introduce a new sample-label contrastive loss to explore the label semantic information and enhance the feature representation learning. For missing labels, we adopt a pseudo-label filling strategy and develop a weighting mechanism to explore the confidently recovered label information. We validate the framework on five standard datasets and the experimental results show that our method achieves superior performance in comparison with the state-of-the-art methods.

RESUMEN

BACKGROUND: Accurate identification of cancer subtypes is crucial for disease prognosis evaluation and personalized patient management. Recent advances in computational methods have demonstrated that multi-omics data provides valuable insights into tumor molecular subtyping. However, the high dimensionality and small sample size of the data may result in ambiguous and overlapping cancer subtypes during clustering. In this study, we propose a novel contrastive-learning-based approach to address this issue. The proposed end-to-end deep learning method can extract crucial information from the multi-omics features by self-supervised learning for patient clustering. RESULTS: By applying our method to nine public cancer datasets, we have demonstrated superior performance compared to existing methods in separating patients with different survival outcomes (p < 0.05). To further evaluate the impact of various omics data on cancer survival, we developed an XGBoost classification model and found that mRNA had the highest importance score, followed by DNA methylation and miRNA. In the presented case study, our method successfully clustered subtypes and identified 14 cancer-related genes, of which 12 (85.7%) were validated through literature review. CONCLUSIONS: Our findings demonstrate that our method is capable of identifying cancer subtypes that are both statistically and biologically significant. The code about COLCS is given at: https://github.com/Mercuriiio/COLCS .

RESUMEN

The precise prediction of hypotension is vital for advancing preemptive patient care strategies. Traditional machine learning approaches, while instrumental in this field, are hampered by their dependence on structured historical data and manual feature extraction techniques. These methods often fall short of recognizing the intricate patterns present in physiological signals. Addressing this limitation, our study introduces an innovative application of deep learning technologies, utilizing a sophisticated end-to-end architecture grounded in XResNet. This architecture is further enhanced by the integration of contrastive learning and a value attention mechanism, specifically tailored to analyze arterial blood pressure (ABP) waveform signals. Our approach improves the performance of hypotension prediction over the existing state-of-theart ABP model [7]. This research represents a step towards optimizing patient care, embodying the next generation of AI-driven healthcare solutions. Through our findings, we demonstrate the promise of deep learning in overcoming the limitations of conventional prediction models, thereby offering an avenue for enhancing patient outcomes in clinical settings.

RESUMEN

Protein Language Models (pLMs) have revolutionized the computational modeling of protein systems, building numerical embeddings that are centered around structural features. To enhance the breadth of biochemically relevant properties available in protein embeddings, we engineered the Annotation Vocabulary, a transformer readable language of protein properties defined by structured ontologies. We trained Annotation Transformers (AT) from the ground up to recover masked protein property inputs without reference to amino acid sequences, building a new numerical feature space on protein descriptions alone. We leverage AT representations in various model architectures, for both protein representation and generation. To showcase the merit of Annotation Vocabulary integration, we performed 515 diverse downstream experiments. Using a novel loss function and only $3 in commercial compute, our premier representation model CAMP produces state-of-the-art embeddings for five out of 15 common datasets with competitive performance on the rest; highlighting the computational efficiency of latent space curation with Annotation Vocabulary. To standardize the comparison of de novo generated protein sequences, we suggest a new sequence alignment-based score that is more flexible and biologically relevant than traditional language modeling metrics. Our generative model, GSM, produces high alignment scores from annotation-only prompts with a BERT-like generation scheme. Of particular note, many GSM hallucinations return statistically significant BLAST hits, where enrichment analysis shows properties matching the annotation prompt - even when the ground truth has low sequence identity to the entire training set. Overall, the Annotation Vocabulary toolbox presents a promising pathway to replace traditional tokens with members of ontologies and knowledge graphs, enhancing transformer models in specific domains. The concise, accurate, and efficient descriptions of proteins by the Annotation Vocabulary offers a novel way to build numerical representations of proteins for protein annotation and design.

RESUMEN

Promoters are important cis-regulatory elements for the regulation of gene expression, and their accurate predictions are crucial for elucidating the biological functions and potential mechanisms of genes. Many previous prokaryotic promoter prediction methods are encouraging in terms of the prediction performance, but most of them focus on the recognition of promoters in only one or a few bacterial species. Moreover, due to ignoring the promoter sequence motifs, the interpretability of predictions with existing methods is limited. In this work, we present a generalized method Prompt (Promoters in multiple prokaryotes) to predict promoters in 16 prokaryotes and improve the interpretability of prediction results. Prompt integrates three methods including RSK (Regression based on Selected k-mer), CL (Contrastive Learning) and MLP (Multilayer Perception), and employs a voting strategy to divide the datasets into high-confidence and low-confidence categories. Results on the promoter prediction tasks in 16 prokaryotes show that the accuracy (Accuracy, Matthews correlation coefficient) of Prompt is greater than 80% in highly credible datasets of 16 prokaryotes, and is greater than 90% in 12 prokaryotes, and Prompt performs the best compared with other existing methods. Moreover, by identifying promoter sequence motifs, Prompt can improve the interpretability of the predictions. Prompt is freely available at https://github.com/duqimeng/PromptPrompt , and will contribute to the research of promoters in prokaryote.

RESUMEN

Deep learning-based 3D/2D surgical navigation registration techniques achieved excellent results. However, these methods are limited by the occlusion of surgical equipment resulting in poor accuracy. We designed a contrastive learning method that treats occluded and unoccluded X-rays as positive samples, maximizing the similarity between the positive samples and reducing interference from occlusion. The designed registration model has Transformer's residual connection (ResTrans), which enhances the long-sequence mapping capability, combined with the contrast learning strategy, ResTrans can adaptively retrieve the valid features in the global range to ensure the performance in the case of occlusion. Further, a learning-based region of interest (RoI) fine-tuning method is designed to refine the misalignment. We conducted experiments on occluded X-rays that contained different surgical devices. The experiment results show that the mean target registration error (mTRE) of ResTrans is 3.25 mm and the running time is 1.59 s. Compared with the state-of-the-art (SOTA) 3D/2D registration methods, our method offers better performance on occluded 3D/2D registration tasks.

Asunto(s)

RESUMEN

Understanding the genetic basis of disease is a fundamental aspect of medical research, as genes are the classic units of heredity and play a crucial role in biological function. Identifying associations between genes and diseases is critical for diagnosis, prevention, prognosis, and drug development. Genes that encode proteins with similar sequences are often implicated in related diseases, as proteins causing identical or similar diseases tend to show limited variation in their sequences. Predicting gene-disease association (GDA) requires time-consuming and expensive experiments on a large number of potential candidate genes. Although methods have been proposed to predict associations between genes and diseases using traditional machine learning algorithms and graph neural networks, these approaches struggle to capture the deep semantic information within the genes and diseases and are dependent on training data. To alleviate this issue, we propose a novel GDA prediction model named FusionGDA, which utilizes a pre-training phase with a fusion module to enrich the gene and disease semantic representations encoded by pre-trained language models. Multi-modal representations are generated by the fusion module, which includes rich semantic information about two heterogeneous biomedical entities: protein sequences and disease descriptions. Subsequently, the pooling aggregation strategy is adopted to compress the dimensions of the multi-modal representation. In addition, FusionGDA employs a pre-training phase leveraging a contrastive learning loss to extract potential gene and disease features by training on a large public GDA dataset. To rigorously evaluate the effectiveness of the FusionGDA model, we conduct comprehensive experiments on five datasets and compare our proposed model with five competitive baseline models on the DisGeNet-Eval dataset. Notably, our case study further demonstrates the ability of FusionGDA to discover hidden associations effectively. The complete code and datasets of our experiments are available at https://github.com/ZhaohanM/FusionGDA.

Asunto(s)

RESUMEN

In recent years, contrastive learning has been a highly favored method for self-supervised representation learning, which significantly improves the unsupervised training of deep image models. Self-supervised learning is a subset of unsupervised learning in which the learning process is supervised by creating pseudolabels from the data themselves. Using supervised final adjustments after unsupervised pretraining is one way to take the most valuable information from a vast collection of unlabeled data and teach from a small number of labeled instances. This study aims firstly to compare contrastive learning with other traditional learning models; secondly to demonstrate by experimental studies the superiority of contrastive learning during classification; thirdly to fine-tune performance using pretrained models and appropriate hyperparameter selection; and finally to address the challenge of using contrastive learning techniques to produce data representations with semantic meaning that are independent of irrelevant factors like position, lighting, and background. Relying on contrastive techniques, the model efficiently captures meaningful representations by discerning similarities and differences between modified copies of the same image. The proposed strategy, involving unsupervised pretraining followed by supervised fine-tuning, improves the robustness, accuracy, and knowledge extraction of deep image models. The results show that even with a modest 5% of data labeled, the semisupervised model achieves an accuracy of 57.72%. However, the use of supervised learning with a contrastive approach and careful hyperparameter tuning increases accuracy to 85.43%. Further adjustment of the hyperparameters resulted in an excellent accuracy of 88.70%.

RESUMEN

Purpose: We introduce a deep learning-based biomarker proposal system for the purpose of accelerating biomarker discovery in age-related macular degeneration (AMD). Design: Retrospective analysis of a large data set of retinal OCT images. Participants: A total of 3456 adults aged between 51 and 102 years whose OCT images were collected under the PINNACLE project. Methods: Our system proposes candidates for novel AMD imaging biomarkers in OCT. It works by first training a neural network using self-supervised contrastive learning to discover, without any clinical annotations, features relating to both known and unknown AMD biomarkers present in 46 496 retinal OCT images. To interpret the learned biomarkers, we partition the images into 30 subsets, termed clusters, that contain similar features. We conduct 2 parallel 1.5-hour semistructured interviews with 2 independent teams of retinal specialists to assign descriptions in clinical language to each cluster. Descriptions of clusters achieving consensus can potentially inform new biomarker candidates. Main Outcome Measures: We checked if each cluster showed clear features comprehensible to retinal specialists, if they related to AMD, and how many described established biomarkers used in grading systems as opposed to recently proposed or potentially new biomarkers. We also compared their prognostic value for late-stage wet and dry AMD against an established clinical grading system and a demographic baseline model. Results: Overall, both teams independently identified clearly distinct characteristics in 27 of 30 clusters, of which 23 were related to AMD. Seven were recognized as known biomarkers used in established grading systems, and 16 depicted biomarker combinations or subtypes that are either not yet used in grading systems, were only recently proposed, or were unknown. Clusters separated incomplete from complete retinal atrophy, intraretinal from subretinal fluid, and thick from thin choroids, and, in simulation, outperformed clinically used grading systems in prognostic value. Conclusions: Using self-supervised deep learning, we were able to automatically propose AMD biomarkers going beyond the set used in clinically established grading systems. Without any clinical annotations, contrastive learning discovered subtle differences between fine-grained biomarkers. Ultimately, we envision that equipping clinicians with discovery-oriented deep learning tools can accelerate the discovery of novel prognostic biomarkers. Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

RESUMEN

Medical image segmentation has made remarkable progress with advances in deep learning technology, depending on the quality and quantity of labeled data. Although various deep learning model structures and training methods have been proposed and high performance has been published, limitations such as inter-class accuracy bias exist in actual clinical applications, especially due to the significant lack of small object performance in multi-organ segmentation tasks. In this paper, we propose an uncertainty-based contrastive learning technique, namely UncerNCE, with an optimal hybrid architecture for high classification and segmentation performance of small organs. Our backbone architecture adopts a hybrid network that employs both convolutional and transformer layers, which have demonstrated remarkable performance in recent years. The key proposal of this study addresses the multi-class accuracy bias and resolves a common tradeoff in existing studies between segmenting regions of small objects and reducing overall noise (i.e., false positives). Uncertainty based contrastive learning based on the proposed hybrid network performs spotlight learning on selected regions based on uncertainty and achieved accurate segmentation for all classes while suppressing noise. Comparison with state-of-the-art techniques demonstrates the superiority of our results on BTCV and 1K data.

Asunto(s)

RESUMEN

Objective.Electroencephalogram (EEG) signals offer invaluable insights into the complexities of emotion generation within the brain. Yet, the variability in EEG signals across individuals presents a formidable obstacle for empirical implementations. Our research addresses these challenges innovatively, focusing on the commonalities within distinct subjects' EEG data.Approach.We introduce a novel approach named Contrastive Learning Graph Convolutional Network (CLGCN). This method captures the distinctive features and crucial channel nodes related to individuals' emotional states. Specifically, CLGCN merges the dual benefits of CL's synchronous multisubject data learning and the GCN's proficiency in deciphering brain connectivity matrices. Understanding multifaceted brain functions and their information interchange processes is realized as CLGCN generates a standardized brain network learning matrix during a dataset's learning process.Main results.Our model underwent rigorous testing on the Database for Emotion Analysis using Physiological Signals (DEAP) and SEED datasets. In the five-fold cross-validation used for dependent subject experimental setting, it achieved an accuracy of 97.13% on the DEAP dataset and surpassed 99% on the SEED and SEED_IV datasets. In the incremental learning experiments with the SEED dataset, merely 5% of the data was sufficient to fine-tune the model, resulting in an accuracy of 92.8% for the new subject. These findings validate the model's efficacy.Significance.This work combines CL with GCN, improving the accuracy of decoding emotional states from EEG signals and offering valuable insights into uncovering the underlying mechanisms of emotional processes in the brain.

Asunto(s)

RESUMEN

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.

Asunto(s)