Functional magnetic resonance imaging (fMRI) is a powerful tool for investigating human brain function. However, the high cost of data acquisition and the inherent subjectivity of psychiatric rating scales often lead to datasets with small sample sizes and variable label quality, especially when targeting a specific neurological condition. Combined with the inherently high dimensionality of fMRI data, these limitations substantially increase the risk of model overfitting. Recent years have seen growing interest in developing fMRI foundation models by combining multiple datasets; however, the computational resources needed for pretraining and fine-tuning are often prohibitive. We show that a lightweight self-supervised framework yields representations that generalize across diverse downstream tasks, outperforming fully supervised baselines and approaching the performance of large-scale models. We introduce BrainSimSiam, a data-efficient self-supervised representation learning framework that leverages positive-only data pairs to learn robust and generalizable features. We demonstrate that the learned representations achieve strong performance across multiple downstream classification and regression tasks, highlighting the potential of BrainSimSiam for data-limited neuroimaging applications. Our implementation is available in https://github.com/Jiyao96/BrainSimSiam-fMRI/.

Recent studies on contrastive learning have achieved remarkable performance solely by leveraging few labels in the context of medical image segmentation. Existing methods mainly focus on instance discrimination and invariant mapping (i.e., pulling positive samples closer and negative samples apart in the feature space). However, they face three common pitfalls: (1) tailness: medical image data usually follows an implicit long-tail class distribution. Blindly leveraging all pixels in training hence can lead to the data imbalance issues, and cause deteriorated performance; (2) consistency: it remains unclear whether a segmentation model has learned meaningful and yet consistent anatomical features due to the intra-class variations between different anatomical features; and (3) diversity: the intra-slice correlations within the entire dataset have received significantly less attention. This motivates us to seek a principled approach for strategically making use of the dataset itself to discover similar yet distinct samples from different anatomical views. In this paper, we introduce a novel semi-supervised 2D medical image segmentation framework termed Mine yOur owNAnatomy (MONA), and make three contributions. First, prior work argues that every pixel equally matters to the model training; we observe empirically that this alone is unlikely to define meaningful anatomical features, mainly due to lacking the supervision signal. We show two simple solutions towards learning invariances-through the use of stronger data augmentations and nearest neighbors. Second, we construct a set of objectives that encourage the model to be capable of decomposing medical images into a collection of anatomical features in an unsupervised manner. Lastly, we both empirically and theoretically, demonstrate the efficacy of our MONA on three benchmark datasets with different labeled settings, achieving new state-of-the-art under different labeled semi-supervised settings. MONA makes minimal assumptions on domain expertise, and hence constitutes a practical and versatile solution in medical image analysis. We provide the PyTorch-like pseudo-code in supplementary.

In whole-body dynamic positron emission tomography (PET), inter-frame subject motion causes spatial misalignment and affects parametric imaging. Many of the current deep learning inter-frame motion correction techniques focus solely on the anatomy-based registration problem, neglecting the tracer kinetics that contains functional information. To directly reduce the Patlak fitting error for 18F-FDG and further improve model performance, we propose an inter-framemotioncorrection framework withPatlak loss optimization integrated into the neural network (MCP-Net). The MCP-Net consists of a multiple-frame motion estimation block, an image-warping block, and an analytical Patlak block that estimates Patlak fitting using motion-corrected frames and the input function. A novel Patlak loss penalty component utilizing mean squared percentage fitting error is added to the loss function to reinforce the motion correction. The parametric images were generated using standard Patlak analysis following motion correction. Our framework enhanced the spatial alignment in both dynamic frames and parametric images and lowered normalized fitting error when compared to both conventional and deep learning benchmarks. MCP-Net also achieved the lowest motion prediction error and showed the best generalization capability. The potential of enhancing network performance and improving the quantitative accuracy of dynamic PET by directly utilizing tracer kinetics is suggested.

Graph Convolutional Neural Networks (GCNs) are widely used for graph analysis. Specifically, in medical applications, GCNs can be used for disease prediction on a population graph, where graph nodes represent individuals and edges represent individual similarities. However, GCNs rely on a vast amount of data, which is challenging to collect for a single medical institution. In addition, a critical challenge that most medical institutions continue to face is addressing disease prediction in isolation with incomplete data information. To address these issues, Federated Learning (FL) allows isolated local institutions to collaboratively train a global model without data sharing. In this work, we propose a framework, FedNI, to leverage network inpainting and inter-institutional data via FL. Specifically, we first federatively train missing node and edge predictor using a graph generative adversarial network (GAN) to complete the missing information of local networks. Then we train a global GCN node classifier across institutions using a federated graph learning platform. The novel design enables us to build more accurate machine learning models by leveraging federated learning and also graph learning approaches. We demonstrate that our federated model outperforms local and baseline FL methods with significant margins on two public neuroimaging datasets.

We propose ACProp (Asynchronous-centering-Prop), an adaptive optimizer which combines centering of second momentum and asynchronous update (e.g. for $t$-th update, denominator uses information up to step $t-1$, while numerator uses gradient at $t$-th step). ACProp has both strong theoretical properties and empirical performance. With the example by Reddi et al. (2018), we show that asynchronous optimizers (e.g. AdaShift, ACProp) have weaker convergence condition than synchronous optimizers (e.g. Adam, RMSProp, AdaBelief); within asynchronous optimizers, we show that centering of second momentum further weakens the convergence condition. We demonstrate that ACProp has a convergence rate of $O(\frac{1}{\sqrt{T}})$ for the stochastic non-convex case, which matches the oracle rate and outperforms the $O(\frac{logT}{\sqrt{T}})$ rate of RMSProp and Adam. We validate ACProp in extensive empirical studies: ACProp outperforms both SGD and other adaptive optimizers in image classification with CNN, and outperforms well-tuned adaptive optimizers in the training of various GAN models, reinforcement learning and transformers. To sum up, ACProp has good theoretical properties including weak convergence condition and optimal convergence rate, and strong empirical performance including good generalization like SGD and training stability like Adam. We provide the implementation at \url{ https://github.com/juntang-zhuang/ACProp-Optimizer}.

Patient motion during dynamic PET imaging can induce errors in myocardial blood flow (MBF) estimation. Motion correction for dynamic cardiac PET is challenging because the rapid tracer kinetics of 82Rb leads to substantial tracer distribution change across different dynamic frames over time, which can cause difficulties for image registration-based motion correction, particularly for early dynamic frames. In this paper, we developed an automatic deep learning-based motion correction (DeepMC) method for dynamic cardiac PET. In this study we focused on the detection and correction of inter-frame rigid translational motion caused by voluntary body movement and pattern change of respiratory motion. A bidirectional-3D LSTM network was developed to fully utilize both local and nonlocal temporal information in the 4D dynamic image data for motion detection. The network was trained and evaluated over motion-free patient scans with simulated motion so that the motion ground-truths are available, where one million samples based on 65 patient scans were used in training, and 600 samples based on 20 patient scans were used in evaluation. The proposed method was also evaluated using additional 10 patient datasets with real motion. We demonstrated that the proposed DeepMC obtained superior performance compared to conventional registration-based methods and other convolutional neural networks (CNN), in terms of motion estimation and MBF quantification accuracy. Once trained, DeepMC is much faster than the registration-based methods and can be easily integrated into the clinical workflow. In the future work, additional investigation is needed to evaluate this approach in a clinical context with realistic patient motion.

The empirical performance of neural ordinary differential equations (NODEs) is significantly inferior to discrete-layer models on benchmark tasks (e.g. image classification). We demonstrate an explanation is the inaccuracy of existing gradient estimation methods: the adjoint method has numerical errors in reverse-mode integration; the naive method suffers from a redundantly deep computation graph. We propose the Adaptive Checkpoint Adjoint (ACA) method: ACA applies a trajectory checkpoint strategy which records the forward- mode trajectory as the reverse-mode trajectory to guarantee accuracy; ACA deletes redundant components for shallow computation graphs; and ACA supports adaptive solvers. On image classification tasks, compared with the adjoint and naive method, ACA achieves half the error rate in half the training time; NODE trained with ACA outperforms ResNet in both accuracy and test-retest reliability. On time-series modeling, ACA outperforms competing methods. Furthermore, NODE with ACA can incorporate physical knowledge to achieve better accuracy.

Most popular optimizers for deep learning can be broadly categorized as adaptive methods (e.g.~Adam) and accelerated schemes (e.g.~stochastic gradient descent (SGD) with momentum). For many models such as convolutional neural networks (CNNs), adaptive methods typically converge faster but generalize worse compared to SGD; for complex settings such as generative adversarial networks (GANs), adaptive methods are typically the default because of their stability. We propose AdaBelief to simultaneously achieve three goals: fast convergence as in adaptive methods, good generalization as in SGD, and training stability. The intuition for AdaBelief is to adapt the stepsize according to the "belief" in the current gradient direction. Viewing the exponential moving average (EMA) of the noisy gradient as the prediction of the gradient at the next time step, if the observed gradient greatly deviates from the prediction, we distrust the current observation and take a small step; if the observed gradient is close to the prediction, we trust it and take a large step. We validate AdaBelief in extensive experiments, showing that it outperforms other methods with fast convergence and high accuracy on image classification and language modeling. Specifically, on ImageNet, AdaBelief achieves comparable accuracy to SGD. Furthermore, in the training of a GAN on Cifar10, AdaBelief demonstrates high stability and improves the quality of generated samples compared to a well-tuned Adam optimizer.

Deep learning models have shown their advantage in many different tasks, including neuroimage analysis. However, to effectively train a high-quality deep learning model, the aggregation of a significant amount of patient information is required. The time and cost for acquisition and annotation in assembling, for example, large fMRI datasets make it difficult to acquire large numbers at a single site. However, due to the need to protect the privacy of patient data, it is hard to assemble a central database from multiple institutions. Federated learning allows for population-level models to be trained without centralizing entities' data by transmitting the global model to local entities, training the model locally, and then averaging the gradients or weights in the global model. However, some studies suggest that private information can be recovered from the model gradients or weights. In this work, we address the problem of multi-site fMRI classification with a privacy-preserving strategy. To solve the problem, we propose a federated learning approach, where a decentralized iterative optimization algorithm is implemented and shared local model weights are altered by a randomization mechanism. Considering the systemic differences of fMRI distributions from different sites, we further propose two domain adaptation methods in this federated learning formulation. We investigate various practical aspects of federated model optimization and compare federated learning with alternative training strategies. Overall, our results demonstrate that it is promising to utilize multi-site data without data sharing to boost neuroimage analysis performance and find reliable disease-related biomarkers. Our proposed pipeline can be generalized to other privacy-sensitive medical data analysis problems. Our code is publicly available at: https://github.com/xxlya/Fed_ABIDE/.

In this project, our goal is to develop a method for interpreting how a neural network makes layer-by-layer embedded decisions when trained for a classification task, and also to use this insight for improving the model performance. To do this, we first approximate the distribution of the image representations in these embeddings using random forest models, the output of which, termed embedding outputs, are used for measuring how the network classifies each sample. Next, we design a pipeline to use this layer embedding output to calibrate the original model output for improved probability calibration and classification. We apply this two-steps method in a fully convolutional neural network trained for a liver tissue classification task on our institutional dataset that contains 20 3D multi-parameter MR images for patients with hepatocellular carcinoma, as well as on a public dataset with 131 3D CT images. The results show that our method is not only able to provide visualizations that are easy to interpret, but that the embedded decision-based information is also useful for improving model performance in terms of probability calibration and classification, achieving the best performance compared to other baseline methods. Moreover, this method is computationally efficient, easy to implement, and robust to hyper-parameters.