This paper presents a comprehensive evaluation of GPT-4V's capabilities across diverse medical imaging tasks, including Radiology Report Generation, Medical Visual Question Answering (VQA), and Visual Grounding. While prior efforts have explored GPT-4V's performance in medical image analysis, to the best of our knowledge, our study represents the first quantitative evaluation on publicly available benchmarks. Our findings highlight GPT-4V's potential in generating descriptive reports for chest X-ray images, particularly when guided by well-structured prompts. Meanwhile, its performance on the MIMIC-CXR dataset benchmark reveals areas for improvement in certain evaluation metrics, such as CIDEr. In the domain of Medical VQA, GPT-4V demonstrates proficiency in distinguishing between question types but falls short of the VQA-RAD benchmark in terms of accuracy. Furthermore, our analysis finds the limitations of conventional evaluation metrics like the BLEU scores, advocating for the development of more semantically robust assessment methods. In the field of Visual Grounding, GPT-4V exhibits preliminary promise in recognizing bounding boxes, but its precision is lacking, especially in identifying specific medical organs and signs. Our evaluation underscores the significant potential of GPT-4V in the medical imaging domain, while also emphasizing the need for targeted refinements to fully unlock its capabilities.
Masked AutoEncoder(MAE) has revolutionized the field of self-supervised learning with its simple yet effective masking and reconstruction strategies. However, despite achieving state-of-the-art performance across various downstream vision tasks, the underlying mechanisms that drive MAE's efficacy are less well-explored compared to the canonical contrastive learning paradigm. In this paper, we explore a new perspective to explain what truly contributes to the "rich hidden representations inside the MAE". Firstly, concerning MAE's generative pretraining pathway, with a unique encoder-decoder architecture to reconstruct images from aggressive masking, we conduct an in-depth analysis of the decoder's behaviors. We empirically find that MAE's decoder mainly learns local features with a limited receptive field, adhering to the well-known Locality Principle. Building upon this locality assumption, we propose a theoretical framework that reformulates the reconstruction-based MAE into a local region-level contrastive learning form for improved understanding. Furthermore, to substantiate the local contrastive nature of MAE, we introduce a Siamese architecture that combines the essence of MAE and contrastive learning without masking and explicit decoder, which sheds light on a unified and more flexible self-supervised learning framework.
Large Language Models (LLMs) have consistently showcased remarkable generalization capabilities when applied to various language tasks. Nonetheless, harnessing the full potential of LLMs for Radiology Report Generation (R2Gen) still presents a challenge, stemming from the inherent disparity in modality between LLMs and the R2Gen task. To bridge this gap effectively, we propose R2GenGPT, which is a novel solution that aligns visual features with the word embedding space of LLMs using an efficient visual alignment module. This innovative approach empowers the previously static LLM to seamlessly integrate and process image information, marking a step forward in optimizing R2Gen performance. R2GenGPT offers the following benefits. First, it attains state-of-the-art (SOTA) performance by training only the lightweight visual alignment module while freezing all the parameters of LLM. Second, it exhibits high training efficiency, as it requires the training of an exceptionally minimal number of parameters while achieving rapid convergence. By employing delta tuning, our model only trains 5M parameters (which constitute just 0.07\% of the total parameter count) to achieve performance close to the SOTA levels. Our code is available at https://github.com/wang-zhanyu/R2GenGPT.
In semi-supervised learning, unlabeled samples can be utilized through augmentation and consistency regularization. However, we observed certain samples, even undergoing strong augmentation, are still correctly classified with high confidence, resulting in a loss close to zero. It indicates that these samples have been already learned well and do not provide any additional optimization benefits to the model. We refer to these samples as ``naive samples". Unfortunately, existing SSL models overlook the characteristics of naive samples, and they just apply the same learning strategy to all samples. To further optimize the SSL model, we emphasize the importance of giving attention to naive samples and augmenting them in a more diverse manner. Sample adaptive augmentation (SAA) is proposed for this stated purpose and consists of two modules: 1) sample selection module; 2) sample augmentation module. Specifically, the sample selection module picks out {naive samples} based on historical training information at each epoch, then the naive samples will be augmented in a more diverse manner in the sample augmentation module. Thanks to the extreme ease of implementation of the above modules, SAA is advantageous for being simple and lightweight. We add SAA on top of FixMatch and FlexMatch respectively, and experiments demonstrate SAA can significantly improve the models. For example, SAA helped improve the accuracy of FixMatch from 92.50% to 94.76% and that of FlexMatch from 95.01% to 95.31% on CIFAR-10 with 40 labels.
Vision Transformer (ViT) architectures are becoming increasingly popular and widely employed to tackle computer vision applications. Their main feature is the capacity to extract global information through the self-attention mechanism, outperforming earlier convolutional neural networks. However, ViT deployment and performance have grown steadily with their size, number of trainable parameters, and operations. Furthermore, self-attention's computational and memory cost quadratically increases with the image resolution. Generally speaking, it is challenging to employ these architectures in real-world applications due to many hardware and environmental restrictions, such as processing and computational capabilities. Therefore, this survey investigates the most efficient methodologies to ensure sub-optimal estimation performances. More in detail, four efficient categories will be analyzed: compact architecture, pruning, knowledge distillation, and quantization strategies. Moreover, a new metric called Efficient Error Rate has been introduced in order to normalize and compare models' features that affect hardware devices at inference time, such as the number of parameters, bits, FLOPs, and model size. Summarizing, this paper firstly mathematically defines the strategies used to make Vision Transformer efficient, describes and discusses state-of-the-art methodologies, and analyzes their performances over different application scenarios. Toward the end of this paper, we also discuss open challenges and promising research directions.
Recent neural networks based surface reconstruction can be roughly divided into two categories, one warping templates explicitly and the other representing 3D surfaces implicitly. To enjoy the advantages of both, we propose a novel 3D representation, Neural Vector Fields (NVF), which adopts the explicit learning process to manipulate meshes and implicit unsigned distance function (UDF) representation to break the barriers in resolution and topology. This is achieved by directly predicting the displacements from surface queries and modeling shapes as Vector Fields, rather than relying on network differentiation to obtain direction fields as most existing UDF-based methods do. In this way, our approach is capable of encoding both the distance and the direction fields so that the calculation of direction fields is differentiation-free, circumventing the non-trivial surface extraction step. Furthermore, building upon NVFs, we propose to incorporate two types of shape codebooks, \ie, NVFs (Lite or Ultra), to promote cross-category reconstruction through encoding cross-object priors. Moreover, we propose a new regularization based on analyzing the zero-curl property of NVFs, and implement this through the fully differentiable framework of our NVF (ultra). We evaluate both NVFs on four surface reconstruction scenarios, including watertight vs non-watertight shapes, category-agnostic reconstruction vs category-unseen reconstruction, category-specific, and cross-domain reconstruction.
Studies on semi-supervised medical image segmentation (SSMIS) have seen fast progress recently. Due to the limited labelled data, SSMIS methods mainly focus on effectively leveraging unlabeled data to enhance the segmentation performance. However, despite their promising performance, current state-of-the-art methods often prioritize integrating complex techniques and loss terms rather than addressing the core challenges of semi-supervised scenarios directly. We argue that the key to SSMIS lies in generating substantial and appropriate prediction disagreement on unlabeled data. To this end, we emphasize the crutiality of data perturbation and model stabilization in semi-supervised segmentation, and propose a simple yet effective approach to boost SSMIS performance significantly, dubbed DPMS. Specifically, we first revisit SSMIS from three distinct perspectives: the data, the model, and the loss, and conduct a comprehensive study of corresponding strategies to examine their effectiveness. Based on these examinations, we then propose DPMS, which adopts a plain teacher-student framework with a standard supervised loss and unsupervised consistency loss. To produce appropriate prediction disagreements, DPMS perturbs the unlabeled data via strong augmentations to enlarge prediction disagreements considerably. On the other hand, using EMA teacher when strong augmentation is applied does not necessarily improve performance. DPMS further utilizes a forwarding-twice and momentum updating strategies for normalization statistics to stabilize the training on unlabeled data effectively. Despite its simplicity, DPMS can obtain new state-of-the-art performance on the public 2D ACDC and 3D LA datasets across various semi-supervised settings, e.g. obtaining a remarkable 22.62% improvement against previous SOTA on ACDC with 5% labels.
To obtain high-quality positron emission tomography (PET) scans while reducing radiation exposure to the human body, various approaches have been proposed to reconstruct standard-dose PET (SPET) images from low-dose PET (LPET) images. One widely adopted technique is the generative adversarial networks (GANs), yet recently, diffusion probabilistic models (DPMs) have emerged as a compelling alternative due to their improved sample quality and higher log-likelihood scores compared to GANs. Despite this, DPMs suffer from two major drawbacks in real clinical settings, i.e., the computationally expensive sampling process and the insufficient preservation of correspondence between the conditioning LPET image and the reconstructed PET (RPET) image. To address the above limitations, this paper presents a coarse-to-fine PET reconstruction framework that consists of a coarse prediction module (CPM) and an iterative refinement module (IRM). The CPM generates a coarse PET image via a deterministic process, and the IRM samples the residual iteratively. By delegating most of the computational overhead to the CPM, the overall sampling speed of our method can be significantly improved. Furthermore, two additional strategies, i.e., an auxiliary guidance strategy and a contrastive diffusion strategy, are proposed and integrated into the reconstruction process, which can enhance the correspondence between the LPET image and the RPET image, further improving clinical reliability. Extensive experiments on two human brain PET datasets demonstrate that our method outperforms the state-of-the-art PET reconstruction methods. The source code is available at \url{https://github.com/Show-han/PET-Reconstruction}.
Semi-supervised learning (SSL) tackles the label missing problem by enabling the effective usage of unlabeled data. While existing SSL methods focus on the traditional setting, a practical and challenging scenario called label Missing Not At Random (MNAR) is usually ignored. In MNAR, the labeled and unlabeled data fall into different class distributions resulting in biased label imputation, which deteriorates the performance of SSL models. In this work, class transition tracking based Pseudo-Rectifying Guidance (PRG) is devised for MNAR. We explore the class-level guidance information obtained by the Markov random walk, which is modeled on a dynamically created graph built over the class tracking matrix. PRG unifies the historical information of class distribution and class transitions caused by the pseudo-rectifying procedure to maintain the model's unbiased enthusiasm towards assigning pseudo-labels to all classes, so as the quality of pseudo-labels on both popular classes and rare classes in MNAR could be improved. Finally, we show the superior performance of PRG across a variety of MNAR scenarios, outperforming the latest SSL approaches combining bias removal solutions by a large margin. Code and model weights are available at https://github.com/NJUyued/PRG4SSL-MNAR.