Medical vision-language models (VLMs) combine computer vision and natural language processing to analyze visual and textual medical data. Our paper reviews recent advancements in developing VLMs specialized for healthcare, focusing on models designed for medical report generation and visual question answering. We provide background on natural language processing and computer vision, explaining how techniques from both fields are integrated into VLMs to enable learning from multimodal data. Key areas we address include the exploration of medical vision-language datasets, in-depth analyses of architectures and pre-training strategies employed in recent noteworthy medical VLMs, and comprehensive discussion on evaluation metrics for assessing VLMs' performance in medical report generation and visual question answering. We also highlight current challenges and propose future directions, including enhancing clinical validity and addressing patient privacy concerns. Overall, our review summarizes recent progress in developing VLMs to harness multimodal medical data for improved healthcare applications.
This study investigated the performance, explainability, and robustness of deployed artificial intelligence (AI) models in predicting mortality during the COVID-19 pandemic and beyond. The first study of its kind, we found that Bayesian Neural Networks (BNNs) and intelligent training techniques allowed our models to maintain performance amidst significant data shifts. Our results emphasize the importance of developing robust AI models capable of matching or surpassing clinician predictions, even under challenging conditions. Our exploration of model explainability revealed that stochastic models generate more diverse and personalized explanations thereby highlighting the need for AI models that provide detailed and individualized insights in real-world clinical settings. Furthermore, we underscored the importance of quantifying uncertainty in AI models which enables clinicians to make better-informed decisions based on reliable predictions. Our study advocates for prioritizing implementation science in AI research for healthcare and ensuring that AI solutions are practical, beneficial, and sustainable in real-world clinical environments. By addressing unique challenges and complexities in healthcare settings, researchers can develop AI models that effectively improve clinical practice and patient outcomes.
The advancements in data acquisition, storage, and processing techniques have resulted in the rapid growth of heterogeneous medical data. Integrating radiological scans, histopathology images, and molecular information with clinical data is essential for developing a holistic understanding of the disease and optimizing treatment. The need for integrating data from multiple sources is further pronounced in complex diseases such as cancer for enabling precision medicine and personalized treatments. This work proposes Multimodal Integration of Oncology Data System (MINDS) - a flexible, scalable, and cost-effective metadata framework for efficiently fusing disparate data from public sources such as the Cancer Research Data Commons (CRDC) into an interconnected, patient-centric framework. MINDS offers an interface for exploring relationships across data types and building cohorts for developing large-scale multimodal machine learning models. By harmonizing multimodal data, MINDS aims to potentially empower researchers with greater analytical ability to uncover diagnostic and prognostic insights and enable evidence-based personalized care. MINDS tracks granular end-to-end data provenance, ensuring reproducibility and transparency. The cloud-native architecture of MINDS can handle exponential data growth in a secure, cost-optimized manner while ensuring substantial storage optimization, replication avoidance, and dynamic access capabilities. Auto-scaling, access controls, and other mechanisms guarantee pipelines' scalability and security. MINDS overcomes the limitations of existing biomedical data silos via an interoperable metadata-driven approach that represents a pivotal step toward the future of oncology data integration.
Deep Neural Networks (DNNs) deployed to the real world are regularly subject to out-of-distribution (OoD) data, various types of noise, and shifting conceptual objectives. This paper proposes a framework for adapting to data distribution drift modeled by benchmark Continual Learning datasets. We develop and evaluate a method of Continual Learning that leverages uncertainty quantification from Bayesian Inference to mitigate catastrophic forgetting. We expand on previous approaches by removing the need for Monte Carlo sampling of the model weights to sample the predictive distribution. We optimize a closed-form Evidence Lower Bound (ELBO) objective approximating the predictive distribution by propagating the first two moments of a distribution, i.e. mean and covariance, through all network layers. Catastrophic forgetting is mitigated by using the closed-form ELBO to approximate the Minimum Description Length (MDL) Principle, inherently penalizing changes in the model likelihood by minimizing the KL Divergence between the variational posterior for the current task and the previous task's variational posterior acting as the prior. Leveraging the approximation of the MDL principle, we aim to initially learn a sparse variational posterior and then minimize additional model complexity learned for subsequent tasks. Our approach is evaluated for the task incremental learning scenario using density propagated versions of fully-connected and convolutional neural networks across multiple sequential benchmark datasets with varying task sequence lengths. Ultimately, this procedure produces a minimally complex network over a series of tasks mitigating catastrophic forgetting.
Historically, the rotorcraft community has experienced a higher fatal accident rate than other aviation segments, including commercial and general aviation. Recent advancements in artificial intelligence (AI) and the application of these technologies in different areas of our lives are both intriguing and encouraging. When developed appropriately for the aviation domain, AI techniques provide an opportunity to help design systems that can address rotorcraft safety challenges. Our recent work demonstrated that AI algorithms could use video data from onboard cameras and correctly identify different flight parameters from cockpit gauges, e.g., indicated airspeed. These AI-based techniques provide a potentially cost-effective solution, especially for small helicopter operators, to record the flight state information and perform post-flight analyses. We also showed that carefully designed and trained AI systems could accurately predict rotorcraft attitude (i.e., pitch and yaw) from outside scenes (images or video data). Ordinary off-the-shelf video cameras were installed inside the rotorcraft cockpit to record the outside scene, including the horizon. The AI algorithm could correctly identify rotorcraft attitude at an accuracy in the range of 80\%. In this work, we combined five different onboard camera viewpoints to improve attitude prediction accuracy to 94\%. In this paper, five onboard camera views included the pilot windshield, co-pilot windshield, pilot Electronic Flight Instrument System (EFIS) display, co-pilot EFIS display, and the attitude indicator gauge. Using video data from each camera view, we trained various convolutional neural networks (CNNs), which achieved prediction accuracy in the range of 79\% % to 90\% %. We subsequently ensembled the learned knowledge from all CNNs and achieved an ensembled accuracy of 93.3\%.
Feature visualization is used to visualize learned features for black box machine learning models. Our approach explores an altered training process to improve interpretability of the visualizations. We argue that by using background removal techniques as a form of robust training, a network is forced to learn more human recognizable features, namely, by focusing on the main object of interest without any distractions from the background. Four different training methods were used to verify this hypothesis. The first used unmodified pictures. The second used a black background. The third utilized Gaussian noise as the background. The fourth approach employed a mix of background removed images and unmodified images. The feature visualization results show that the background removed images reveal a significant improvement over the baseline model. These new results displayed easily recognizable features from their respective classes, unlike the model trained on unmodified data.
In the last few years, many works have tried to explain the predictions of deep learning models. Few methods, however, have been proposed to verify the accuracy or faithfulness of these explanations. Recently, influence functions, which is a method that approximates the effect that leave-one-out training has on the loss function, has been shown to be fragile. The proposed reason for their fragility remains unclear. Although previous work suggests the use of regularization to increase robustness, this does not hold in all cases. In this work, we seek to investigate the experiments performed in the prior work in an effort to understand the underlying mechanisms of influence function fragility. First, we verify influence functions using procedures from the literature under conditions where the convexity assumptions of influence functions are met. Then, we relax these assumptions and study the effects of non-convexity by using deeper models and more complex datasets. Here, we analyze the key metrics and procedures that are used to validate influence functions. Our results indicate that the validation procedures may cause the observed fragility.
The expansion of explainable artificial intelligence as a field of research has generated numerous methods of visualizing and understanding the black box of a machine learning model. Attribution maps are generally used to highlight the parts of the input image that influence the model to make a specific decision. On the other hand, the robustness of machine learning models to natural noise and adversarial attacks is also being actively explored. This paper focuses on evaluating methods of attribution mapping to find whether robust neural networks are more explainable. We explore this problem within the application of classification for medical imaging. Explainability research is at an impasse. There are many methods of attribution mapping, but no current consensus on how to evaluate them and determine the ones that are the best. Our experiments on multiple datasets (natural and medical imaging) and various attribution methods reveal that two popular evaluation metrics, Deletion and Insertion, have inherent limitations and yield contradictory results. We propose a new explainability faithfulness metric (called EvalAttAI) that addresses the limitations of prior metrics. Using our novel evaluation, we found that Bayesian deep neural networks using the Variational Density Propagation technique were consistently more explainable when used with the best performing attribution method, the Vanilla Gradient. However, in general, various types of robust neural networks may not be more explainable, despite these models producing more visually plausible attribution maps.
Cancer has relational information residing at varying scales, modalities, and resolutions of the acquired data, such as radiology, pathology, genomics, proteomics, and clinical records. Integrating diverse data types can improve the accuracy and reliability of cancer diagnosis and treatment. There can be disease-related information that is too subtle for humans or existing technological tools to discern visually. Traditional methods typically focus on partial or unimodal information about biological systems at individual scales and fail to encapsulate the complete spectrum of the heterogeneous nature of data. Deep neural networks have facilitated the development of sophisticated multimodal data fusion approaches that can extract and integrate relevant information from multiple sources. Recent deep learning frameworks such as Graph Neural Networks (GNNs) and Transformers have shown remarkable success in multimodal learning. This review article provides an in-depth analysis of the state-of-the-art in GNNs and Transformers for multimodal data fusion in oncology settings, highlighting notable research studies and their findings. We also discuss the foundations of multimodal learning, inherent challenges, and opportunities for integrative learning in oncology. By examining the current state and potential future developments of multimodal data integration in oncology, we aim to demonstrate the promising role that multimodal neural networks can play in cancer prevention, early detection, and treatment through informed oncology practices in personalized settings.
The quality and richness of feature maps extracted by convolution neural networks (CNNs) and vision Transformers (ViTs) directly relate to the robust model performance. In medical computer vision, these information-rich features are crucial for detecting rare cases within large datasets. This work presents the "Scopeformer," a novel multi-CNN-ViT model for intracranial hemorrhage classification in computed tomography (CT) images. The Scopeformer architecture is scalable and modular, which allows utilizing various CNN architectures as the backbone with diversified output features and pre-training strategies. We propose effective feature projection methods to reduce redundancies among CNN-generated features and to control the input size of ViTs. Extensive experiments with various Scopeformer models show that the model performance is proportional to the number of convolutional blocks employed in the feature extractor. Using multiple strategies, including diversifying the pre-training paradigms for CNNs, different pre-training datasets, and style transfer techniques, we demonstrate an overall improvement in the model performance at various computational budgets. Later, we propose smaller compute-efficient Scopeformer versions with three different types of input and output ViT configurations. Efficient Scopeformers use four different pre-trained CNN architectures as feature extractors to increase feature richness. Our best Efficient Scopeformer model achieved an accuracy of 96.94\% and a weighted logarithmic loss of 0.083 with an eight times reduction in the number of trainable parameters compared to the base Scopeformer. Another version of the Efficient Scopeformer model further reduced the parameter space by almost 17 times with negligible performance reduction. Hybrid CNNs and ViTs might provide the desired feature richness for developing accurate medical computer vision models