Xiao Fang, Yi Lin, Dong Zhang, Kwang-Ting Cheng, Hao Chen
Pre-trained large vision-language models (VLMs) like CLIP have revolutionized visual representation learning using natural language as supervisions, and demonstrated promising generalization ability. In this work, we propose ViP, a novel visual symptom-guided prompt learning framework for medical image analysis, which facilitates general knowledge transfer from CLIP. ViP consists of two key components: a visual symptom generator (VSG) and a dual-prompt network. Specifically, VSG aims to extract explicable visual symptoms from pre-trained large language models, while the dual-prompt network utilizes these visual symptoms to guide the training on two learnable prompt modules, i.e., context prompt and merge prompt, which effectively adapts our framework to medical image analysis via large VLMs. Extensive experimental results demonstrate that ViP can outperform state-of-the-art methods on two challenging datasets.
Shuhan Li, Dong Zhang, Xiaomeng Li, Chubin Ou, Lin An, Yanwu Xu, Kwang-Ting Cheng
Optical Coherence Tomography Angiography (OCTA) is a crucial tool in the clinical screening of retinal diseases, allowing for accurate 3D imaging of blood vessels through non-invasive scanning. However, the hardware-based approach for acquiring OCTA images presents challenges due to the need for specialized sensors and expensive devices. In this paper, we introduce a novel method called TransPro, which can translate the readily available 3D Optical Coherence Tomography (OCT) images into 3D OCTA images without requiring any additional hardware modifications. Our TransPro method is primarily driven by two novel ideas that have been overlooked by prior work. The first idea is derived from a critical observation that the OCTA projection map is generated by averaging pixel values from its corresponding B-scans along the Z-axis. Hence, we introduce a hybrid architecture incorporating a 3D adversarial generative network and a novel Heuristic Contextual Guidance (HCG) module, which effectively maintains the consistency of the generated OCTA images between 3D volumes and projection maps. The second idea is to improve the vessel quality in the translated OCTA projection maps. As a result, we propose a novel Vessel Promoted Guidance (VPG) module to enhance the attention of network on retinal vessels. Experimental results on two datasets demonstrate that our TransPro outperforms state-of-the-art approaches, with relative improvements around 11.4% in MAE, 2.7% in PSNR, 2% in SSIM, 40% in VDE, and 9.1% in VDC compared to the baseline method. The code is available at: https://github.com/ustlsh/TransPro.
Jiabo Ma, Zhengrui Guo, Fengtao Zhou, Yihui Wang, Yingxue Xu, Jinbang Li, Fang Yan, Yu Cai, Zhengjie Zhu, Cheng Jin, Yi Lin, Xinrui Jiang, Chenglong Zhao, Danyi Li, Anjia Han, Zhenhui Li, Ronald Cheong Kin Chan, Jiguang Wang, Peng Fei, Kwang-Ting Cheng, Shaoting Zhang, Li Liang, Hao Chen
Foundation models pretrained on large-scale datasets are revolutionizing the field of computational pathology (CPath). The generalization ability of foundation models is crucial for the success in various downstream clinical tasks. However, current foundation models have only been evaluated on a limited type and number of tasks, leaving their generalization ability and overall performance unclear. To address this gap, we established a most comprehensive benchmark to evaluate the performance of off-the-shelf foundation models across six distinct clinical task types, encompassing a total of 72 specific tasks, including slide-level classification, survival prediction, ROI-tissue classification, ROI retrieval, visual question answering, and report generation. Our findings reveal that existing foundation models excel at certain task types but struggle to effectively handle the full breadth of clinical tasks. To improve the generalization of pathology foundation models, we propose a unified knowledge distillation framework consisting of both expert and self-knowledge distillation, where the former allows the model to learn from the knowledge of multiple expert models, while the latter leverages self-distillation to enable image representation learning via local-global alignment. Based on this framework, we curated a dataset of 96,000 whole slide images (WSIs) and developed a Generalizable Pathology Foundation Model (GPFM). This advanced model was trained on a substantial dataset comprising 190 million images extracted from approximately 72,000 publicly available slides, encompassing 34 major tissue types. Evaluated on the established benchmark, GPFM achieves an impressive average rank of 1.6, with 42 tasks ranked 1st, while the second-best model, UNI, attains an average rank of 3.7, with only 6 tasks ranked 1st.
Yu Cai, Weiwen Zhang, Hao Chen, Kwang-Ting Cheng
Anomaly detection (AD) aims at detecting abnormal samples that deviate from the expected normal patterns. Generally, it can be trained merely on normal data, without a requirement for abnormal samples, and thereby plays an important role in rare disease recognition and health screening in the medical domain. Despite the emergence of numerous methods for medical AD, the lack of a fair and comprehensive evaluation causes ambiguous conclusions and hinders the development of this field. To address this problem, this paper builds a benchmark with unified comparison. Seven medical datasets with five image modalities, including chest X-rays, brain MRIs, retinal fundus images, dermatoscopic images, and histopathology images, are curated for extensive evaluation. Thirty typical AD methods, including reconstruction and self-supervised learning-based methods, are involved in comparison of image-level anomaly classification and pixel-level anomaly segmentation. Furthermore, for the first time, we systematically investigate the effect of key components in existing methods, revealing unresolved challenges and potential future directions. The datasets and code are available at https://github.com/caiyu6666/MedIAnomaly.
Xixi Jiang, Chen Yang, Dong Zhang, Pingcheng Dong, Xin Yang, Kwang-Ting Cheng
Vision Transformer models have shown impressive effectiveness in the surgical video understanding tasks through long-range dependency modeling. However, current methods suffer from prohibitive computational costs due to processing massive spatiotemporal tokens across video frames. While prior work on token merging has advanced model efficiency, they fail to adequately consider the inherent spatiotemporal structure of video data and overlook the heterogeneous nature of information distribution, leading to suboptimal performance. In this paper, we propose a spatiotemporal information mining token merging (STIM-TM) method, representing the first dedicated approach for surgical video understanding. STIM-TM introduces a decoupled strategy that reduces token redundancy along temporal and spatial dimensions independently. Specifically, the temporal component merges spatially corresponding tokens from consecutive frames using saliency weighting, preserving critical sequential information and maintaining continuity. Meanwhile, the spatial component prioritizes merging static tokens through temporal stability analysis, protecting dynamic regions containing essential surgical information. Operating in a training-free manner, STIM-TM achieves significant efficiency gains with over $65\%$ GFLOPs reduction while preserving competitive accuracy across comprehensive surgical video tasks. Our method also supports efficient training of long-sequence surgical videos, addressing computational bottlenecks in surgical applications.
Zhiqiang Shen, Zechun Liu, Dejia Xu, Zitian Chen, Kwang-Ting Cheng, Marios Savvides
This work aims to empirically clarify a recently discovered perspective that label smoothing is incompatible with knowledge distillation. We begin by introducing the motivation behind on how this incompatibility is raised, i.e., label smoothing erases relative information between teacher logits. We provide a novel connection on how label smoothing affects distributions of semantically similar and dissimilar classes. Then we propose a metric to quantitatively measure the degree of erased information in sample's representation. After that, we study its one-sidedness and imperfection of the incompatibility view through massive analyses, visualizations and comprehensive experiments on Image Classification, Binary Networks, and Neural Machine Translation. Finally, we broadly discuss several circumstances wherein label smoothing will indeed lose its effectiveness. Project page: http://zhiqiangshen.com/projects/LS_and_KD/index.html.
Yu Cai, Hao Chen, Xin Yang, Yu Zhou, Kwang-Ting Cheng
Chest X-ray (CXR) is the most typical radiological exam for diagnosis of various diseases. Due to the expensive and time-consuming annotations, detecting anomalies in CXRs in an unsupervised fashion is very promising. However, almost all of the existing methods consider anomaly detection as a one-class classification (OCC) problem. They model the distribution of only known normal images during training and identify the samples not conforming to normal profile as anomalies in the testing phase. A large number of unlabeled images containing anomalies are thus ignored in the training phase, although they are easy to obtain in clinical practice. In this paper, we propose a novel strategy, Dual-distribution Discrepancy for Anomaly Detection (DDAD), utilizing both known normal images and unlabeled images. The proposed method consists of two modules. During training, one module takes both known normal and unlabeled images as inputs, capturing anomalous features from unlabeled images in some way, while the other one models the distribution of only known normal images. Subsequently, inter-discrepancy between the two modules, and intra-discrepancy inside the module that is trained on only normal images are designed as anomaly scores to indicate anomalies. Experiments on three CXR datasets demonstrate that the proposed DDAD achieves consistent, significant gains and outperforms state-of-the-art methods. Code is available at https://github.com/caiyu6666/DDAD.
Xijie Huang, Jeffry Wicaksana, Shichao Li, Kwang-Ting Cheng
The growth in the aging population requires caregivers to improve both efficiency and quality of healthcare. In this study, we develop an automatic, vision-based system for monitoring and analyzing the physical and mental well-being of senior citizens. Through collaboration with Haven of Hope Christian Service, we collect video recording data in the care center with surveillance cameras. We then process and extract personalized facial, activity, and interaction features from the video data using deep neural networks. This integrated health information systems can assist caregivers to gain better insights into the seniors they are taking care of. These insights, including wellness metrics and long-term health patterns of senior citizens, can help caregivers update their caregiving strategies. We report the findings of our analysis and evaluate the system quantitatively. We also summarize technical challenges and additional functionalities and technologies needed for offering a comprehensive system.
Zechun Liu, Zhiqiang Shen, Yun Long, Eric Xing, Kwang-Ting Cheng, Chas Leichner
This paper aims to explore the feasibility of neural architecture search (NAS) given only a pre-trained model without using any original training data. This is an important circumstance for privacy protection, bias avoidance, etc., in real-world scenarios. To achieve this, we start by synthesizing usable data through recovering the knowledge from a pre-trained deep neural network. Then we use the synthesized data and their predicted soft-labels to guide neural architecture search. We identify that the NAS task requires the synthesized data (we target at image domain here) with enough semantics, diversity, and a minimal domain gap from the natural images. For semantics, we propose recursive label calibration to produce more informative outputs. For diversity, we propose a regional update strategy to generate more diverse and semantically-enriched synthetic data. For minimal domain gap, we use input and feature-level regularization to mimic the original data distribution in latent space. We instantiate our proposed framework with three popular NAS algorithms: DARTS, ProxylessNAS and SPOS. Surprisingly, our results demonstrate that the architectures discovered by searching with our synthetic data achieve accuracy that is comparable to, or even higher than, architectures discovered by searching from the original ones, for the first time, deriving the conclusion that NAS can be done effectively with no need of access to the original or called natural data if the synthesis method is well designed.
Xian Lin, Li Yu, Kwang-Ting Cheng, Zengqiang Yan
Objective: Transformers, born to remedy the inadequate receptive fields of CNNs, have drawn explosive attention recently. However, the daunting computational complexity of global representation learning, together with rigid window partitioning, hinders their deployment in medical image segmentation. This work aims to address the above two issues in transformers for better medical image segmentation. Methods: We propose a boundary-aware lightweight transformer (BATFormer) that can build cross-scale global interaction with lower computational complexity and generate windows flexibly under the guidance of entropy. Specifically, to fully explore the benefits of transformers in long-range dependency establishment, a cross-scale global transformer (CGT) module is introduced to jointly utilize multiple small-scale feature maps for richer global features with lower computational complexity. Given the importance of shape modeling in medical image segmentation, a boundary-aware local transformer (BLT) module is constructed. Different from rigid window partitioning in vanilla transformers which would produce boundary distortion, BLT adopts an adaptive window partitioning scheme under the guidance of entropy for both computational complexity reduction and shape preservation. Results: BATFormer achieves the best performance in Dice of 92.84%, 91.97%, 90.26%, and 96.30% for the average, right ventricle, myocardium, and left ventricle respectively on the ACDC dataset and the best performance in Dice, IoU, and ACC of 90.76%, 84.64%, and 96.76% respectively on the ISIC 2018 dataset. More importantly, BATFormer requires the least amount of model parameters and the lowest computational complexity compared to the state-of-the-art approaches. Conclusion and Significance: Our results demonstrate the necessity of developing customized transformers for efficient and better medical image segmentation.
Shichao Li, Zechun Liu, Zhiqiang Shen, Kwang-Ting Cheng
We propose a new object-centric framework for learning-based stereo 3D object detection. Previous studies build scene-centric representations that do not consider the significant variation among outdoor instances and thus lack the flexibility and functionalities that an instance-level model can offer. We build such an instance-level model by formulating and tackling a local update problem, i.e., how to predict a refined update given an initial 3D cuboid guess. We demonstrate how solving this problem can complement scene-centric approaches in (i) building a coarse-to-fine multi-resolution system, (ii) performing model-agnostic object location refinement, and (iii) conducting stereo 3D tracking-by-detection. Extensive experiments demonstrate the effectiveness of our approach, which achieves state-of-the-art performance on the KITTI benchmark. Code and pre-trained models are available at https://github.com/Nicholasli1995/SNVC.
Weihang Dai, Xiaomeng Li, Xinpeng Ding, Kwang-Ting Cheng
Left-ventricular ejection fraction (LVEF) is an important indicator of heart failure. Existing methods for LVEF estimation from video require large amounts of annotated data to achieve high performance, e.g. using 10,030 labeled echocardiogram videos to achieve mean absolute error (MAE) of 4.10. Labeling these videos is time-consuming however and limits potential downstream applications to other heart diseases. This paper presents the first semi-supervised approach for LVEF prediction. Unlike general video prediction tasks, LVEF prediction is specifically related to changes in the left ventricle (LV) in echocardiogram videos. By incorporating knowledge learned from predicting LV segmentations into LVEF regression, we can provide additional context to the model for better predictions. To this end, we propose a novel Cyclical Self-Supervision (CSS) method for learning video-based LV segmentation, which is motivated by the observation that the heartbeat is a cyclical process with temporal repetition. Prediction masks from our segmentation model can then be used as additional input for LVEF regression to provide spatial context for the LV region. We also introduce teacher-student distillation to distill the information from LV segmentation masks into an end-to-end LVEF regression model that only requires video inputs. Results show our method outperforms alternative semi-supervised methods and can achieve MAE of 4.17, which is competitive with state-of-the-art supervised performance, using half the number of labels. Validation on an external dataset also shows improved generalization ability from using our method. Our code is available at https://github.com/xmed-lab/CSS-SemiVideo.
Yu Cai, Hao Chen, Xin Yang, Yu Zhou, Kwang-Ting Cheng
Medical anomaly detection is a crucial yet challenging task aimed at recognizing abnormal images to assist in diagnosis. Due to the high-cost annotations of abnormal images, most methods utilize only known normal images during training and identify samples deviating from the normal profile as anomalies in the testing phase. Many readily available unlabeled images containing anomalies are thus ignored in the training phase, restricting the performance. To solve this problem, we introduce one-class semi-supervised learning (OC-SSL) to utilize known normal and unlabeled images for training, and propose Dual-distribution Discrepancy for Anomaly Detection (DDAD) based on this setting. Ensembles of reconstruction networks are designed to model the distribution of normal images and the distribution of both normal and unlabeled images, deriving the normative distribution module (NDM) and unknown distribution module (UDM). Subsequently, the intra-discrepancy of NDM and inter-discrepancy between the two modules are designed as anomaly scores. Furthermore, we propose a new perspective on self-supervised learning, which is designed to refine the anomaly scores rather than detect anomalies directly. Five medical datasets, including chest X-rays, brain MRIs and retinal fundus images, are organized as benchmarks for evaluation. Experiments on these benchmarks comprehensively compare a wide range of anomaly detection methods and demonstrate that our method achieves significant gains and outperforms the state-of-the-art. Code and organized benchmarks are available at https://github.com/caiyu6666/DDAD-ASR.
Jichang Yang, Hegan Chen, Jia Chen, Songqi Wang, Shaocong Wang, Yifei Yu, Xi Chen, Bo Wang, Xinyuan Zhang, Binbin Cui, Yi Li, Ning Lin, Meng Xu, Yi Li, Xiaoxin Xu, Xiaojuan Qi, Zhongrui Wang, Xumeng Zhang, Dashan Shang, Han Wang, Qi Liu, Kwang-Ting Cheng, Ming Liu
Human brains image complicated scenes when reading a novel. Replicating this imagination is one of the ultimate goals of AI-Generated Content (AIGC). However, current AIGC methods, such as score-based diffusion, are still deficient in terms of rapidity and efficiency. This deficiency is rooted in the difference between the brain and digital computers. Digital computers have physically separated storage and processing units, resulting in frequent data transfers during iterative calculations, incurring large time and energy overheads. This issue is further intensified by the conversion of inherently continuous and analog generation dynamics, which can be formulated by neural differential equations, into discrete and digital operations. Inspired by the brain, we propose a time-continuous and analog in-memory neural differential equation solver for score-based diffusion, employing emerging resistive memory. The integration of storage and computation within resistive memory synapses surmount the von Neumann bottleneck, benefiting the generative speed and energy efficiency. The closed-loop feedback integrator is time-continuous, analog, and compact, physically implementing an infinite-depth neural network. Moreover, the software-hardware co-design is intrinsically robust to analog noise. We experimentally validate our solution with 180 nm resistive memory in-memory computing macros. Demonstrating equivalent generative quality to the software baseline, our system achieved remarkable enhancements in generative speed for both unconditional and conditional generation tasks, by factors of 64.8 and 156.5, respectively. Moreover, it accomplished reductions in energy consumption by factors of 5.2 and 4.1. Our approach heralds a new horizon for hardware solutions in edge computing for generative AI applications.
Weihang Dai, Xiaomeng Li, Taihui Yu, Di Zhao, Jun Shen, Kwang-Ting Cheng
Atrial Fibrillation (AF) is characterized by rapid, irregular heartbeats, and can lead to fatal complications such as heart failure. The disease is divided into two sub-types based on severity, which can be automatically classified through CT volumes for disease screening of severe cases. However, existing classification approaches rely on generic radiomic features that may not be optimal for the task, whilst deep learning methods tend to over-fit to the high-dimensional volume inputs. In this work, we propose a novel radiomics-informed deep-learning method, RIDL, that combines the advantages of deep learning and radiomic approaches to improve AF sub-type classification. Unlike existing hybrid techniques that mostly rely on naïve feature concatenation, we observe that radiomic feature selection methods can serve as an information prior, and propose supplementing low-level deep neural network (DNN) features with locally computed radiomic features. This reduces DNN over-fitting and allows local variations between radiomic features to be better captured. Furthermore, we ensure complementary information is learned by deep and radiomic features by designing a novel feature de-correlation loss. Combined, our method addresses the limitations of deep learning and radiomic approaches and outperforms state-of-the-art radiomic, deep learning, and hybrid approaches, achieving 86.9% AUC for the AF sub-type classification task. Code is available at https://github.com/xmed-lab/RIDL.
Shih-yang Liu, Zechun Liu, Xijie Huang, Pingcheng Dong, Kwang-Ting Cheng
We propose LLM-FP4 for quantizing both weights and activations in large language models (LLMs) down to 4-bit floating-point values, in a post-training manner. Existing post-training quantization (PTQ) solutions are primarily integer-based and struggle with bit widths below 8 bits. Compared to integer quantization, floating-point (FP) quantization is more flexible and can better handle long-tail or bell-shaped distributions, and it has emerged as a default choice in many hardware platforms. One characteristic of FP quantization is that its performance largely depends on the choice of exponent bits and clipping range. In this regard, we construct a strong FP-PTQ baseline by searching for the optimal quantization parameters. Furthermore, we observe a high inter-channel variance and low intra-channel variance pattern in activation distributions, which adds activation quantization difficulty. We recognize this pattern to be consistent across a spectrum of transformer models designed for diverse tasks, such as LLMs, BERT, and Vision Transformer models. To tackle this, we propose per-channel activation quantization and show that these additional scaling factors can be reparameterized as exponential biases of weights, incurring a negligible cost. Our method, for the first time, can quantize both weights and activations in the LLaMA-13B to only 4-bit and achieves an average score of 63.1 on the common sense zero-shot reasoning tasks, which is only 5.8 lower than the full-precision model, significantly outperforming the previous state-of-the-art by 12.7 points. Code is available at: https://github.com/nbasyl/LLM-FP4.
Jeffry Wicaksana, Zengqiang Yan, Kwang-Ting Cheng
Limited training data and severe class imbalance pose significant challenges to developing clinically robust deep learning models. Federated learning (FL) addresses the former by enabling different medical clients to collaboratively train a deep model without sharing privacy-sensitive data. However, class imbalance worsens due to variation in inter-client class distribution. We propose federated learning with non-parametric regularization (FedNPR and FedNPR-Per, a personalized version of FedNPR) to regularize the feature extractor and enhance useful and discriminative signal in the feature space. Our extensive experiments show that FedNPR outperform the existing state-of-the art FL approaches in class imbalance skin lesion classification and intracranial hemorrhage identification. Additionally, the non-parametric regularization module consistently improves the performance of existing state-of-the-art FL approaches. We believe that NPR is a valuable tool in FL under clinical settings.
Shichao Li, Xijie Huang, Zechun Liu, Kwang-Ting Cheng
We present a new learning-based framework S-3D-RCNN that can recover accurate object orientation in SO(3) and simultaneously predict implicit rigid shapes from stereo RGB images. For orientation estimation, in contrast to previous studies that map local appearance to observation angles, we propose a progressive approach by extracting meaningful Intermediate Geometrical Representations (IGRs). This approach features a deep model that transforms perceived intensities from one or two views to object part coordinates to achieve direct egocentric object orientation estimation in the camera coordinate system. To further achieve finer description inside 3D bounding boxes, we investigate the implicit shape estimation problem from stereo images. We model visible object surfaces by designing a point-based representation, augmenting IGRs to explicitly address the unseen surface hallucination problem. Extensive experiments validate the effectiveness of the proposed IGRs, and S-3D-RCNN achieves superior 3D scene understanding performance. We also designed new metrics on the KITTI benchmark for our evaluation of implicit shape estimation.
Dong Zhang, Kwang-Ting Cheng
Thanks to the recent achievements in task-driven image quality enhancement (IQE) models like ESTR, the image enhancement model and the visual recognition model can mutually enhance each other's quantitation while producing high-quality processed images that are perceivable by our human vision systems. However, existing task-driven IQE models tend to overlook an underlying fact -- different levels of vision tasks have varying and sometimes conflicting requirements of image features. To address this problem, this paper proposes a generalized gradient promotion (GradProm) training strategy for task-driven IQE of medical images. Specifically, we partition a task-driven IQE system into two sub-models, i.e., a mainstream model for image enhancement and an auxiliary model for visual recognition. During training, GradProm updates only parameters of the image enhancement model using gradients of the visual recognition model and the image enhancement model, but only when gradients of these two sub-models are aligned in the same direction, which is measured by their cosine similarity. In case gradients of these two sub-models are not in the same direction, GradProm only uses the gradient of the image enhancement model to update its parameters. Theoretically, we have proved that the optimization direction of the image enhancement model will not be biased by the auxiliary visual recognition model under the implementation of GradProm. Empirically, extensive experimental results on four public yet challenging medical image datasets demonstrated the superior performance of GradProm over existing state-of-the-art methods.
Jiajun Zhou, Jiajun Wu, Yizhao Gao, Yuhao Ding, Chaofan Tao, Boyu Li, Fengbin Tu, Kwang-Ting Cheng, Hayden Kwok-Hay So, Ngai Wong
To accelerate the inference of deep neural networks (DNNs), quantization with low-bitwidth numbers is actively researched. A prominent challenge is to quantize the DNN models into low-bitwidth numbers without significant accuracy degradation, especially at very low bitwidths (< 8 bits). This work targets an adaptive data representation with variable-length encoding called DyBit. DyBit can dynamically adjust the precision and range of separate bit-field to be adapted to the DNN weights/activations distribution. We also propose a hardware-aware quantization framework with a mixed-precision accelerator to trade-off the inference accuracy and speedup. Experimental results demonstrate that the inference accuracy via DyBit is 1.997% higher than the state-of-the-art at 4-bit quantization, and the proposed framework can achieve up to 8.1x speedup compared with the original model.