Mask-Guided Vision Transformer for Few-Shot Learning.

Learning with little data is challenging but often inevitable in various application scenarios where the labeled data are limited and costly. Recently, few-shot learning (FSL) gained increasing attention because of its generalizability of prior knowledge to new tasks that contain only a few samples. However, for data-intensive models such as vision transformer (ViT), current fine-tuning-based FSL approaches are inefficient in knowledge generalization and, thus, degenerate the downstream task performances. In this article, we propose a novel mask-guided ViT (MG-ViT) to achieve an effective and efficient FSL on the ViT model. The key idea is to apply a mask on image patches to screen out the task-irrelevant ones and to guide the ViT focusing on task-relevant and discriminative patches during FSL. Particularly, MG-ViT only introduces an additional mask operation and a residual connection, enabling the inheritance of parameters from pretrained ViT without any other cost. To optimally select representative few-shot samples, we also include an active learning-based sample selection method to further improve the generalizability of MG-ViT-based FSL. We evaluate the proposed MG-ViT on classification, object detection, and segmentation tasks using gradient-weighted class activation mapping (Grad-CAM) to generate masks. The experimental results show that the MG-ViT model significantly improves the performance and efficiency compared with general fine-tuning-based ViT and ResNet models, providing novel insights and a concrete approach toward generalizing data-intensive and large-scale deep learning models for FSL.

Hierarchical functional differences between gyri and sulci at different scales.

Gyri and sulci are 2 fundamental cortical folding patterns of the human brain. Recent studies have suggested that gyri and sulci may play different functional roles given their structural and functional heterogeneity. However, our understanding of the functional differences between gyri and sulci remains limited due to several factors. Firstly, previous studies have typically focused on either the spatial or temporal domain, neglecting the inherently spatiotemporal nature of brain functions. Secondly, analyses have often been restricted to either local or global scales, leaving the question of hierarchical functional differences unresolved. Lastly, there has been a lack of appropriate analytical tools for interpreting the hierarchical spatiotemporal features that could provide insights into these differences. To overcome these limitations, in this paper, we proposed a novel hierarchical interpretable autoencoder (HIAE) to explore the hierarchical functional difference between gyri and sulci. Central to our approach is its capability to extract hierarchical features via a deep convolutional autoencoder and then to map these features into an embedding vector using a carefully designed feature interpreter. This process transforms the features into interpretable spatiotemporal patterns, which are pivotal in investigating the functional disparities between gyri and sulci. We evaluate the proposed framework on Human Connectome Project task functional magnetic resonance imaging dataset. The experiments demonstrate that the HIAE model can effectively extract and interpret hierarchical spatiotemporal features that are neuroscientifically meaningful. The analyses based on the interpreted features suggest that gyri are more globally activated, whereas sulci are more locally activated, demonstrating a distinct transition in activation patterns as the scale shifts from local to global. Overall, our study provides novel insights into the brain's anatomy-function relationship.

Technical note: Generalizable and promptable artificial intelligence model to augment clinical delineation in radiation oncology.

BACKGROUND: Efficient and accurate delineation of organs at risk (OARs) is a critical procedure for treatment planning and dose evaluation. Deep learning-based auto-segmentation of OARs has shown promising results and is increasingly being used in radiation therapy. However, existing deep learning-based auto-segmentation approaches face two challenges in clinical practice: generalizability and human-AI interaction. A generalizable and promptable auto-segmentation model, which segments OARs of multiple disease sites simultaneously and supports on-the-fly human-AI interaction, can significantly enhance the efficiency of radiation therapy treatment planning. PURPOSE: Meta's segment anything model (SAM) was proposed as a generalizable and promptable model for next-generation natural image segmentation. We further evaluated the performance of SAM in radiotherapy segmentation. METHODS: Computed tomography (CT) images of clinical cases from four disease sites at our institute were collected: prostate, lung, gastrointestinal, and head & neck. For each case, we selected the OARs important in radiotherapy treatment planning. We then compared both the Dice coefficients and Jaccard indices derived from three distinct methods: manual delineation (ground truth), automatic segmentation using SAM's 'segment anything' mode, and automatic segmentation using SAM's 'box prompt' mode that implements manual interaction via live prompts during segmentation. RESULTS: Our results indicate that SAM's segment anything mode can achieve clinically acceptable segmentation results in most OARs with Dice scores higher than 0.7. SAM's box prompt mode further improves Dice scores by 0.1∼0.5. Similar results were observed for Jaccard indices. The results show that SAM performs better for prostate and lung, but worse for gastrointestinal and head & neck. When considering the size of organs and the distinctiveness of their boundaries, SAM shows better performance for large organs with distinct boundaries, such as lung and liver, and worse for smaller organs with less distinct boundaries, like parotid and cochlea. CONCLUSIONS: Our results demonstrate SAM's robust generalizability with consistent accuracy in automatic segmentation for radiotherapy. Furthermore, the advanced box-prompt method enables the users to augment auto-segmentation interactively and dynamically, leading to patient-specific auto-segmentation in radiation therapy. SAM's generalizability across different disease sites and different modalities makes it feasible to develop a generic auto-segmentation model in radiotherapy.

Differentiating ChatGPT-Generated and Human-Written Medical Texts: Quantitative Study.

BACKGROUND: Large language models, such as ChatGPT, are capable of generating grammatically perfect and human-like text content, and a large number of ChatGPT-generated texts have appeared on the internet. However, medical texts, such as clinical notes and diagnoses, require rigorous validation, and erroneous medical content generated by ChatGPT could potentially lead to disinformation that poses significant harm to health care and the general public. OBJECTIVE: This study is among the first on responsible artificial intelligence-generated content in medicine. We focus on analyzing the differences between medical texts written by human experts and those generated by ChatGPT and designing machine learning workflows to effectively detect and differentiate medical texts generated by ChatGPT. METHODS: We first constructed a suite of data sets containing medical texts written by human experts and generated by ChatGPT. We analyzed the linguistic features of these 2 types of content and uncovered differences in vocabulary, parts-of-speech, dependency, sentiment, perplexity, and other aspects. Finally, we designed and implemented machine learning methods to detect medical text generated by ChatGPT. The data and code used in this paper are published on GitHub. RESULTS: Medical texts written by humans were more concrete, more diverse, and typically contained more useful information, while medical texts generated by ChatGPT paid more attention to fluency and logic and usually expressed general terminologies rather than effective information specific to the context of the problem. A bidirectional encoder representations from transformers-based model effectively detected medical texts generated by ChatGPT, and the F1 score exceeded 95%. CONCLUSIONS: Although text generated by ChatGPT is grammatically perfect and human-like, the linguistic characteristics of generated medical texts were different from those written by human experts. Medical text generated by ChatGPT could be effectively detected by the proposed machine learning algorithms. This study provides a pathway toward trustworthy and accountable use of large language models in medicine.

A generic framework for embedding human brain function with temporally correlated autoencoder.

Learning an effective and compact representation of human brain function from high-dimensional fMRI data is crucial for studying the brain's functional organization. Traditional representation methods such as independent component analysis (ICA) and sparse dictionary learning (SDL) mainly rely on matrix decomposition which represents the brain function as spatial brain networks and the corresponding temporal patterns. The correspondence of those brain networks across individuals are built by viewing them as one-hot vectors and then performing the matching. However, those one-hot vectors do not encode the regularity and/or variability of different brains very well, and thus are limited in effectively representing the functional brain activities across individuals and among different time points. To address this problem, in this paper, we formulate the human brain functional representation as an embedding problem, and propose a novel embedding framework based on the Transformer model to encode the brain function in a compact, stereotyped and comparable latent space where the brain activities are represented as dense embedding vectors. We evaluate the proposed embedding framework on the publicly available Human Connectome Project (HCP) task fMRI dataset. The experiments on brain state prediction task indicate the effectiveness and generalizability of the learned embedding. We also explore the interpretability of the learned embedding from both spatial and temporal perspective. In general, our approach provides novel insights on representing the regularity and variability of human brain function in a general, comparable, and stereotyped latent space.

Core-Periphery Principle Guided Redesign of Self-Attention in Transformers.

Designing more efficient, reliable, and explainable neural network architectures is critical to studies that are based on artificial intelligence (AI) techniques. Previous studies, by post-hoc analysis, have found that the best-performing ANNs surprisingly resemble biological neural networks (BNN), which indicates that ANNs and BNNs may share some common principles to achieve optimal performance in either machine learning or cognitive/behavior tasks. Inspired by this phenomenon, we proactively instill organizational principles of BNNs to guide the redesign of ANNs. We leverage the Core-Periphery (CP) organization, which is widely found in human brain networks, to guide the information communication mechanism in the self-attention of vision transformer (ViT) and name this novel framework as CP-ViT. In CP-ViT, the attention operation between nodes is defined by a sparse graph with a Core-Periphery structure (CP graph), where the core nodes are redesigned and reorganized to play an integrative role and serve as a center for other periphery nodes to exchange information. We evaluated the proposed CP-ViT on multiple public datasets, including medical image datasets (INbreast) and natural image datasets. Interestingly, by incorporating the BNN-derived principle (CP structure) into the redesign of ViT, our CP-ViT outperforms other state-of-the-art ANNs. In general, our work advances the state of the art in three aspects: 1) This work provides novel insights for brain-inspired AI: we can utilize the principles found in BNNs to guide and improve our ANN architecture design; 2) We show that there exist sweet spots of CP graphs that lead to CP-ViTs with significantly improved performance; and 3) The core nodes in CP-ViT correspond to task-related meaningful and important image patches, which can significantly enhance the interpretability of the trained deep model.

Artificial general intelligence for radiation oncology.

The emergence of artificial general intelligence (AGI) is transforming radiation oncology. As prominent vanguards of AGI, large language models (LLMs) such as GPT-4 and PaLM 2 can process extensive texts and large vision models (LVMs) such as the Segment Anything Model (SAM) can process extensive imaging data to enhance the efficiency and precision of radiation therapy. This paper explores full-spectrum applications of AGI across radiation oncology including initial consultation, simulation, treatment planning, treatment delivery, treatment verification, and patient follow-up. The fusion of vision data with LLMs also creates powerful multimodal models that elucidate nuanced clinical patterns. Together, AGI promises to catalyze a shift towards data-driven, personalized radiation therapy. However, these models should complement human expertise and care. This paper provides an overview of how AGI can transform radiation oncology to elevate the standard of patient care in radiation oncology, with the key insight being AGI's ability to exploit multimodal clinical data at scale.

Surviving ChatGPT in healthcare.

At the dawn of of Artificial General Intelligence (AGI), the emergence of large language models such as ChatGPT show promise in revolutionizing healthcare by improving patient care, expanding medical access, and optimizing clinical processes. However, their integration into healthcare systems requires careful consideration of potential risks, such as inaccurate medical advice, patient privacy violations, the creation of falsified documents or images, overreliance on AGI in medical education, and the perpetuation of biases. It is crucial to implement proper oversight and regulation to address these risks, ensuring the safe and effective incorporation of AGI technologies into healthcare systems. By acknowledging and mitigating these challenges, AGI can be harnessed to enhance patient care, medical knowledge, and healthcare processes, ultimately benefiting society as a whole.

Survey on natural language processing in medical image analysis. / 自然语言处理在医学影像分析中的应用.

Recent advancement in natural language processing (NLP) and medical imaging empowers the wide applicability of deep learning models. These developments have increased not only data understanding, but also knowledge of state-of-the-art architectures and their real-world potentials. Medical imaging researchers have recognized the limitations of only targeting images, as well as the importance of integrating multimodal inputs into medical image analysis. The lack of comprehensive surveys of the current literature, however, impedes the progress of this domain. Existing research perspectives, as well as the architectures, tasks, datasets, and performance measures examined in the present literature, are reviewed in this work, and we also provide a brief description of possible future directions in the field, aiming to provide researchers and healthcare professionals with a detailed summary of existing academic research and to provide rational insights to facilitate future research.

Hierarchical Organization of Functional Brain Networks Revealed by Hybrid Spatiotemporal Deep Learning.

Hierarchical organization of brain function has been an established concept in the neuroscience field for a long time, however, it has been rarely demonstrated how such hierarchical macroscale functional networks are actually organized in the human brain. In this study, to answer this question, we propose a novel methodology to provide an evidence of hierarchical organization of functional brain networks. This article introduces the hybrid spatiotemporal deep learning (HSDL), by jointly using deep belief networks (DBNs) and deep least absolute shrinkage and selection operator (LASSO) to reveal the temporal hierarchical features and spatial hierarchical maps of brain networks based on the Human Connectome Project 900 functional magnetic resonance imaging (fMRI) data sets. Briefly, the key idea of HSDL is to extract the weights between two adjacent layers of DBNs, which are then treated as the hierarchical dictionaries for deep LASSO to identify the corresponding hierarchical spatial maps. Our results demonstrate that both spatial and temporal aspects of dozens of functional networks exhibit multiscale properties that can be well characterized and interpreted based on existing computational tools and neuroscience knowledge. Our proposed novel hybrid deep model is used to provide the first insightful opportunity to reveal the potential hierarchical organization of time series and functional brain networks, using task-based fMRI signals of human brain.