Hongyi Pan

Debesh Jha, Nikhil Kumar Tomar, Vanshali Sharma et al.

Colonoscopy is the primary method for examination, detection, and removal of polyps. However, challenges such as variations among the endoscopists' skills, bowel quality preparation, and the complex nature of the large intestine contribute to high polyp miss-rate. These missed polyps can develop into cancer later, underscoring the importance of improving the detection methods. To address this gap of lack of publicly available, multi-center large and diverse datasets for developing automatic methods for polyp detection and segmentation, we introduce PolypDB, a large scale publicly available dataset that contains 3934 still polyp images and their corresponding ground truth from real colonoscopy videos. PolypDB comprises images from five modalities: Blue Light Imaging (BLI), Flexible Imaging Color Enhancement (FICE), Linked Color Imaging (LCI), Narrow Band Imaging (NBI), and White Light Imaging (WLI) from three medical centers in Norway, Sweden, and Vietnam. We provide a benchmark on each modality and center, including federated learning settings using popular segmentation and detection benchmarks. PolypDB is public and can be downloaded at \url{https://osf.io/pr7ms/}. More information about the dataset, segmentation, detection, federated learning benchmark and train-test split can be found at \url{https://github.com/DebeshJha/PolypDB}.

Hongyi Pan, Emadeldeen Hamdan, Xin Zhu et al.

In this paper, we propose a set of transform-based neural network layers as an alternative to the $3\times3$ Conv2D layers in Convolutional Neural Networks (CNNs). The proposed layers can be implemented based on orthogonal transforms such as the Discrete Cosine Transform (DCT), Hadamard transform (HT), and biorthogonal Block Wavelet Transform (BWT). Furthermore, by taking advantage of the convolution theorems, convolutional filtering operations are performed in the transform domain using element-wise multiplications. Trainable soft-thresholding layers, that remove noise in the transform domain, bring nonlinearity to the transform domain layers. Compared to the Conv2D layer, which is spatial-agnostic and channel-specific, the proposed layers are location-specific and channel-specific. Moreover, these proposed layers reduce the number of parameters and multiplications significantly while improving the accuracy results of regular ResNets on the ImageNet-1K classification task. Furthermore, they can be inserted with a batch normalization layer before the global average pooling layer in the conventional ResNets as an additional layer to improve classification accuracy.

Hongyi Pan, Bin Wang, Zheyuan Zhang et al.

Domain generalization aims to train models on multiple source domains so that they can generalize well to unseen target domains. Among many domain generalization methods, Fourier-transform-based domain generalization methods have gained popularity primarily because they exploit the power of Fourier transformation to capture essential patterns and regularities in the data, making the model more robust to domain shifts. The mainstream Fourier-transform-based domain generalization swaps the Fourier amplitude spectrum while preserving the phase spectrum between the source and the target images. However, it neglects background interference in the amplitude spectrum. To overcome this limitation, we introduce a soft-thresholding function in the Fourier domain. We apply this newly designed algorithm to retinal fundus image segmentation, which is important for diagnosing ocular diseases but the neural network's performance can degrade across different sources due to domain shifts. The proposed technique basically enhances fundus image augmentation by eliminating small values in the Fourier domain and providing better generalization. The innovative nature of the soft thresholding fused with Fourier-transform-based domain generalization improves neural network models' performance by reducing the target images' background interference significantly. Experiments on public data validate our approach's effectiveness over conventional and state-of-the-art methods with superior segmentation metrics.

Hongyi Pan, Xin Zhu, Salih Atici et al.

In this paper, we propose a novel Discrete Cosine Transform (DCT)-based neural network layer which we call DCT-perceptron to replace the $3\times3$ Conv2D layers in the Residual neural Network (ResNet). Convolutional filtering operations are performed in the DCT domain using element-wise multiplications by taking advantage of the Fourier and DCT Convolution theorems. A trainable soft-thresholding layer is used as the nonlinearity in the DCT perceptron. Compared to ResNet's Conv2D layer which is spatial-agnostic and channel-specific, the proposed layer is location-specific and channel-specific. The DCT-perceptron layer reduces the number of parameters and multiplications significantly while maintaining comparable accuracy results of regular ResNets in CIFAR-10 and ImageNet-1K. Moreover, the DCT-perceptron layer can be inserted with a batch normalization layer before the global average pooling layer in the conventional ResNets as an additional layer to improve classification accuracy.

Hongyi Pan, Salih Atici, Ahmet Enis Cetin

In this paper, we introduce a convolutional network which we call MultiPodNet consisting of a combination of two or more convolutional networks which process the input image in parallel to achieve the same goal. Output feature maps of parallel convolutional networks are fused at the fully connected layer of the network. We experimentally observed that three parallel pod networks (TripodNet) produce the best results in commonly used object recognition datasets. Baseline pod networks can be of any type. In this paper, we use ResNets as baseline networks and their inputs are augmented image patches. The number of parameters of the TripodNet is about three times that of a single ResNet. We train the TripodNet using the standard backpropagation type algorithms. In each individual ResNet, parameters are initialized with different random numbers during training. The TripodNet achieved state-of-the-art performance on CIFAR-10 and ImageNet datasets. For example, it improved the accuracy of a single ResNet from 91.66% to 92.47% under the same training process on the CIFAR-10 dataset.

Yalda Zafari, Hongyi Pan, Gorkem Durak et al.

The development of clinically reliable artificial intelligence (AI) systems for mammography is hindered by profound heterogeneity in data quality, metadata standards, and population distributions across public datasets. This heterogeneity introduces dataset-specific biases that severely compromise the generalizability of the model, a fundamental barrier to clinical deployment. We present MammoClean, a public framework for standardization and bias quantification in mammography datasets. MammoClean standardizes case selection, image processing (including laterality and intensity correction), and unifies metadata into a consistent multi-view structure. We provide a comprehensive review of breast anatomy, imaging characteristics, and public mammography datasets to systematically identify key sources of bias. Applying MammoClean to three heterogeneous datasets (CBIS-DDSM, TOMPEI-CMMD, VinDr-Mammo), we quantify substantial distributional shifts in breast density and abnormality prevalence. Critically, we demonstrate the direct impact of data corruption: AI models trained on corrupted datasets exhibit significant performance degradation compared to their curated counterparts. By using MammoClean to identify and mitigate bias sources, researchers can construct unified multi-dataset training corpora that enable development of robust models with superior cross-domain generalization. MammoClean provides an essential, reproducible pipeline for bias-aware AI development in mammography, facilitating fairer comparisons and advancing the creation of safe, effective systems that perform equitably across diverse patient populations and clinical settings. The open-source code is publicly available from: https://github.com/Minds-R-Lab/MammoClean.

Salih Atici, Hongyi Pan, Ahmet Enis Cetin

In this paper, we propose a novel optimization algorithm for training machine learning models called Input Normalized Stochastic Gradient Descent (INSGD), inspired by the Normalized Least Mean Squares (NLMS) algorithm used in adaptive filtering. When training complex models on large datasets, the choice of optimizer parameters, particularly the learning rate, is crucial to avoid divergence. Our algorithm updates the network weights using stochastic gradient descent with $\ell_1$ and $\ell_2$-based normalizations applied to the learning rate, similar to NLMS. However, unlike existing normalization methods, we exclude the error term from the normalization process and instead normalize the update term using the input vector to the neuron. Our experiments demonstrate that our optimization algorithm achieves higher accuracy levels compared to different initialization settings. We evaluate the efficiency of our training algorithm on benchmark datasets using ResNet-18, WResNet-20, ResNet-50, and a toy neural network. Our INSGD algorithm improves the accuracy of ResNet-18 on CIFAR-10 from 92.42\% to 92.71\%, WResNet-20 on CIFAR-100 from 76.20\% to 77.39\%, and ResNet-50 on ImageNet-1K from 75.52\% to 75.67\%.

Xin Zhu, Daoguang Yang, Hongyi Pan et al.

The lack of an efficient compression model remains a challenge for the wireless transmission of gearbox data in non-contact gear fault diagnosis problems. In this paper, we present a signal-adaptive asymmetrical autoencoder with a transform domain layer to compress sensor signals. First, a new discrete cosine Stockwell transform (DCST) layer is introduced to replace linear layers in a multi-layer autoencoder. A trainable filter is implemented in the DCST domain by utilizing the multiplication property of the convolution. A trainable hard-thresholding layer is applied to reduce redundant data in the DCST layer to make the feature map sparse. In comparison to the linear layer, the DCST layer reduces the number of trainable parameters and improves the accuracy of data reconstruction. Second, training the autoencoder with a sparsifying DCST layer only requires a small number of datasets. The proposed method is superior to other autoencoder-based methods on the University of Connecticut (UoC) and Southeast University (SEU) gearbox datasets, as the average quality score is improved by 2.00% at the lowest and 32.35% at the highest with a limited number of training samples

Ziliang Hong, Emadeldeen Hamdan, Yifei Zhao et al.

This paper surveys different publicly available neural network models used for detecting wildfires using regular visible-range cameras which are placed on hilltops or forest lookout towers. The neural network models are pre-trained on ImageNet-1K and fine-tuned on a custom wildfire dataset. The performance of these models is evaluated on a diverse set of wildfire images, and the survey provides useful information for those interested in using transfer learning for wildfire detection. Swin Transformer-tiny has the highest AUC value but ConvNext-tiny detects all the wildfire events and has the lowest false alarm rate in our dataset.

Nastaran Darabi, Maeesha Binte Hashem, Hongyi Pan et al.

The edge processing of deep neural networks (DNNs) is becoming increasingly important due to its ability to extract valuable information directly at the data source to minimize latency and energy consumption. Frequency-domain model compression, such as with the Walsh-Hadamard transform (WHT), has been identified as an efficient alternative. However, the benefits of frequency-domain processing are often offset by the increased multiply-accumulate (MAC) operations required. This paper proposes a novel approach to an energy-efficient acceleration of frequency-domain neural networks by utilizing analog-domain frequency-based tensor transformations. Our approach offers unique opportunities to enhance computational efficiency, resulting in several high-level advantages, including array micro-architecture with parallelism, ADC/DAC-free analog computations, and increased output sparsity. Our approach achieves more compact cells by eliminating the need for trainable parameters in the transformation matrix. Moreover, our novel array micro-architecture enables adaptive stitching of cells column-wise and row-wise, thereby facilitating perfect parallelism in computations. Additionally, our scheme enables ADC/DAC-free computations by training against highly quantized matrix-vector products, leveraging the parameter-free nature of matrix multiplications. Another crucial aspect of our design is its ability to handle signed-bit processing for frequency-based transformations. This leads to increased output sparsity and reduced digitization workload. On a 16$\times$16 crossbars, for 8-bit input processing, the proposed approach achieves the energy efficiency of 1602 tera operations per second per Watt (TOPS/W) without early termination strategy and 5311 TOPS/W with early termination strategy at VDD = 0.8 V.

Xin Zhu, Hongyi Pan, Shuaiang Rong et al.

Electroencephalogram (EEG) data compression is necessary for wireless recording applications to reduce the amount of data that needs to be transmitted. In this paper, an asymmetrical sparse autoencoder with a discrete cosine transform (DCT) layer is proposed to compress EEG signals. The encoder module of the autoencoder has a combination of a fully connected linear layer and the DCT layer to reduce redundant data using hard-thresholding nonlinearity. Furthermore, the DCT layer includes trainable hard-thresholding parameters and scaling layers to give emphasis or de-emphasis on individual DCT coefficients. Finally, the one-by-one convolutional layer generates the latent space. The sparsity penalty-based cost function is employed to keep the feature map as sparse as possible in the latent space. The latent space data is transmitted to the receiver. The decoder module of the autoencoder is designed using the inverse DCT and two fully connected linear layers to improve the accuracy of data reconstruction. In comparison to other state-of-the-art methods, the proposed method significantly improves the average quality score in various data compression experiments.

Emadeldeen Hamdan, Hongyi Pan, Ahmet Enis Cetin

Structured state space models' (SSMs) development in recent studies, such as Mamba and Mamba2, outperformed and solved the computational inefficiency of transformers and large language models at small to medium scale. In this work, we introduce the concept of controllability and observability to the original Mamba SSM's architecture in our Sparse-Mamba (S-Mamba) for natural language processing (NLP) applications. Moreover, we reinforce stability on the $nxn$ $A$ matrix on Mmaba2. The Mamba SSMs architecture drops the need for attention layers or multilayer perception blocks in transformers. However, current Mamba models lack reinforcement of controllability in state-space equations for computing the $A$, $B$, $C$, and $D$ matrices at each time step, leading to increased complexity and computational costs. Furthermore, the $A$ matrix in Mamba2 is not always stable. We demonstrate a reduction of parameters compared to the first published Mamba and Mamba2. We showcase an improvement in perplexity by 5\% and a decrease in training time by 3\% after reinforcing controllability and observability on the original Mamba architecture in our proposed S-Mamba. We further enforce stability on the $A$ matrix in Mamba2 to improve the loss and perplexity of the model. The controllable and stable $n \times n$ state matrix $A$ is sparse, and it has only $n$ free parameters. Our novel approach will ensure controllable/observable and stable SSMs, which will be the gate key for Mamba3.

Hongyi Pan, Gorkem Durak, Halil Ertugrul Aktas et al.

Publicly available full-field digital mammography (FFDM) datasets remain limited in size, clinical labels, and vendor diversity, which hinders the training of robust models. We present LUMINA, a curated, multi-vendor FFDM dataset that explicitly encodes acquisition energy and vendor metadata to expose clinically relevant appearance shifts that current benchmarks overlook. This innovative resource comprises 1824 images from 468 patients (960 benign, 864 malignant) with pathology-confirmed outcomes, BI-RADS assessments, and breast-density annotations. LUMINA spans six acquisition systems and both high- and low-energy styles, exposing vendor- and energy-driven appearance shifts. To reduce cross-vendor/energy drift while preserving lesion morphology, we introduce a foreground-only, pixel-space alignment (''energy harmonization'') that aligns each image to a low-energy reference style, leaving the zero-valued background unchanged. By benchmarking modern CNN and transformer baselines on three clinically meaningful tasks -- diagnosis (benign vs. malignant), BI-RADS risk grouping, and density -- we unify single-vs-two-view evaluation and show that two-view models consistently outperform single-view; in our benchmark, EfficientNet-B0 attains AUC 93.54% for diagnosis, and Swin-T yields the best macro-AUC 89.43% for density. Harmonization improves AUC/ACC across backbones and yields more focal Grad-CAM localization around suspicious regions. Being a richly annotated resource, LUMINA thus provides (a) a vendor-diverse, energy-labeled benchmark and (b) a model-agnostic harmonization protocol that together catalyze reliable, deployable mammography AI.

Zheyuan Zhang, Monica Dou, Linkai Peng et al.

Advertisement videos serve as a rich and valuable source of purpose-driven information, encompassing high-quality visual, textual, and contextual cues designed to engage viewers. They are often more complex than general videos of similar duration due to their structured narratives and rapid scene transitions, posing significant challenges to multi-modal large language models (MLLMs). In this work, we introduce VideoAds, the first dataset tailored for benchmarking the performance of MLLMs on advertisement videos. VideoAds comprises well-curated advertisement videos with complex temporal structures, accompanied by \textbf{manually} annotated diverse questions across three core tasks: visual finding, video summary, and visual reasoning. We propose a quantitative measure to compare VideoAds against existing benchmarks in terms of video complexity. Through extensive experiments, we find that Qwen2.5-VL-72B, an opensource MLLM, achieves 73.35\% accuracy on VideoAds, outperforming GPT-4o (66.82\%) and Gemini-1.5 Pro (69.66\%); the two proprietary models especially fall behind the opensource model in video summarization and reasoning, but perform the best in visual finding. Notably, human experts easily achieve a remarkable accuracy of 94.27\%. These results underscore the necessity of advancing MLLMs' temporal modeling capabilities and highlight VideoAds as a potentially pivotal benchmark for future research in understanding video that requires high FPS sampling. The dataset and evaluation code will be publicly available at https://videoadsbenchmark.netlify.app.

Zheyuan Zhang, Elif Keles, Gorkem Durak et al.

Automated volumetric segmentation of the pancreas on cross-sectional imaging is needed for diagnosis and follow-up of pancreatic diseases. While CT-based pancreatic segmentation is more established, MRI-based segmentation methods are understudied, largely due to a lack of publicly available datasets, benchmarking research efforts, and domain-specific deep learning methods. In this retrospective study, we collected a large dataset (767 scans from 499 participants) of T1-weighted (T1W) and T2-weighted (T2W) abdominal MRI series from five centers between March 2004 and November 2022. We also collected CT scans of 1,350 patients from publicly available sources for benchmarking purposes. We developed a new pancreas segmentation method, called PanSegNet, combining the strengths of nnUNet and a Transformer network with a new linear attention module enabling volumetric computation. We tested PanSegNet's accuracy in cross-modality (a total of 2,117 scans) and cross-center settings with Dice and Hausdorff distance (HD95) evaluation metrics. We used Cohen's kappa statistics for intra and inter-rater agreement evaluation and paired t-tests for volume and Dice comparisons, respectively. For segmentation accuracy, we achieved Dice coefficients of 88.3% (std: 7.2%, at case level) with CT, 85.0% (std: 7.9%) with T1W MRI, and 86.3% (std: 6.4%) with T2W MRI. There was a high correlation for pancreas volume prediction with R^2 of 0.91, 0.84, and 0.85 for CT, T1W, and T2W, respectively. We found moderate inter-observer (0.624 and 0.638 for T1W and T2W MRI, respectively) and high intra-observer agreement scores. All MRI data is made available at https://osf.io/kysnj/. Our source code is available at https://github.com/NUBagciLab/PaNSegNet.

Bin Wang, Hongyi Pan, Armstrong Aboah et al.

Eye tracking research is important in computer vision because it can help us understand how humans interact with the visual world. Specifically for high-risk applications, such as in medical imaging, eye tracking can help us to comprehend how radiologists and other medical professionals search, analyze, and interpret images for diagnostic and clinical purposes. Hence, the application of eye tracking techniques in disease classification has become increasingly popular in recent years. Contemporary works usually transform gaze information collected by eye tracking devices into visual attention maps (VAMs) to supervise the learning process. However, this is a time-consuming preprocessing step, which stops us from applying eye tracking to radiologists' daily work. To solve this problem, we propose a novel gaze-guided graph neural network (GNN), GazeGNN, to leverage raw eye-gaze data without being converted into VAMs. In GazeGNN, to directly integrate eye gaze into image classification, we create a unified representation graph that models both images and gaze pattern information. With this benefit, we develop a real-time, real-world, end-to-end disease classification algorithm for the first time in the literature. This achievement demonstrates the practicality and feasibility of integrating real-time eye tracking techniques into the daily work of radiologists. To our best knowledge, GazeGNN is the first work that adopts GNN to integrate image and eye-gaze data. Our experiments on the public chest X-ray dataset show that our proposed method exhibits the best classification performance compared to existing methods. The code is available at https://github.com/ukaukaaaa/GazeGNN.

Songlin Xu, Hao-Ning Wen, Hongyi Pan et al.

Student simulation supports educators to improve teaching by interacting with virtual students. However, most existing approaches ignore the modulation effects of course materials because of two challenges: the lack of datasets with granularly annotated course materials, and the limitation of existing simulation models in processing extremely long textual data. To solve the challenges, we first run a 6-week education workshop from N = 60 students to collect fine-grained data using a custom built online education system, which logs students' learning behaviors as they interact with lecture materials over time. Second, we propose a transferable iterative reflection (TIR) module that augments both prompting-based and finetuning-based large language models (LLMs) for simulating learning behaviors. Our comprehensive experiments show that TIR enables the LLMs to perform more accurate student simulation than classical deep learning models, even with limited demonstration data. Our TIR approach better captures the granular dynamism of learning performance and inter-student correlations in classrooms, paving the way towards a ''digital twin'' for online education.

Hongyi Pan, Ziliang Hong, Gorkem Durak et al.

Accurate classification of Intraductal Papillary Mucinous Neoplasms (IPMN) is essential for identifying high-risk cases that require timely intervention. In this study, we develop a federated learning framework for multi-center IPMN classification utilizing a comprehensive pancreas MRI dataset. This dataset includes 652 T1-weighted and 655 T2-weighted MRI images, accompanied by corresponding IPMN risk scores from 7 leading medical institutions, making it the largest and most diverse dataset for IPMN classification to date. We assess the performance of DenseNet-121 in both centralized and federated settings for training on distributed data. Our results demonstrate that the federated learning approach achieves high classification accuracy comparable to centralized learning while ensuring data privacy across institutions. This work marks a significant advancement in collaborative IPMN classification, facilitating secure and high-accuracy model training across multiple centers.

Hongyi Pan, Gorkem Durak, Zheyuan Zhang et al.

Federated learning (FL) enables collaborative model training across institutions without sharing sensitive data, making it an attractive solution for medical imaging tasks. However, traditional FL methods, such as Federated Averaging (FedAvg), face difficulties in generalizing across domains due to variations in imaging protocols and patient demographics across institutions. This challenge is particularly evident in pancreas MRI segmentation, where anatomical variability and imaging artifacts significantly impact performance. In this paper, we conduct a comprehensive evaluation of FL algorithms for pancreas MRI segmentation and introduce a novel approach that incorporates adaptive aggregation weights. By dynamically adjusting the contribution of each client during model aggregation, our method accounts for domain-specific differences and improves generalization across heterogeneous datasets. Experimental results demonstrate that our approach enhances segmentation accuracy and reduces the impact of domain shift compared to conventional FL methods while maintaining privacy-preserving capabilities. Significant performance improvements are observed across multiple hospitals (centers).

Debesh Jha, Nikhil Kumar Tomar, Koushik Biswas et al.

Accurate segmentation of organs from abdominal CT scans is essential for clinical applications such as diagnosis, treatment planning, and patient monitoring. To handle challenges of heterogeneity in organ shapes, sizes, and complex anatomical relationships, we propose a \textbf{\textit{\ac{MDNet}}}, an encoder-decoder network that uses the pre-trained \textit{MiT-B2} as the encoder and multiple different decoder networks. Each decoder network is connected to a different part of the encoder via a multi-scale feature enhancement dilated block. With each decoder, we increase the depth of the network iteratively and refine segmentation masks, enriching feature maps by integrating previous decoders' feature maps. To refine the feature map further, we also utilize the predicted masks from the previous decoder to the current decoder to provide spatial attention across foreground and background regions. MDNet effectively refines the segmentation mask with a high dice similarity coefficient (DSC) of 0.9013 and 0.9169 on the Liver Tumor segmentation (LiTS) and MSD Spleen datasets. Additionally, it reduces Hausdorff distance (HD) to 3.79 for the LiTS dataset and 2.26 for the spleen segmentation dataset, underscoring the precision of MDNet in capturing the complex contours. Moreover, \textit{\ac{MDNet}} is more interpretable and robust compared to the other baseline models.

Xin Zhu, Hongyi Pan, Yury Velichko et al.

Magnetic field inhomogeneity correction remains a challenging task in MRI analysis. Most established techniques are designed for brain MRI by supposing that image intensities in the identical tissue follow a uniform distribution. Such an assumption cannot be easily applied to other organs, especially those that are small in size and heterogeneous in texture (large variations in intensity), such as the prostate. To address this problem, this paper proposes a probabilistic Hadamard U-Net (PHU-Net) for prostate MRI bias field correction. First, a novel Hadamard U-Net (HU-Net) is introduced to extract the low-frequency scalar field, multiplied by the original input to obtain the prototypical corrected image. HU-Net converts the input image from the time domain into the frequency domain via Hadamard transform. In the frequency domain, high-frequency components are eliminated using the trainable filter (scaling layer), hard-thresholding layer, and sparsity penalty. Next, a conditional variational autoencoder is used to encode possible bias field-corrected variants into a low-dimensional latent space. Random samples drawn from latent space are then incorporated with a prototypical corrected image to generate multiple plausible images. Experimental results demonstrate the effectiveness of PHU-Net in correcting bias-field in prostate MRI with a fast inference speed. It has also been shown that prostate MRI segmentation accuracy improves with the high-quality corrected images from PHU-Net. The code will be available in the final version of this manuscript.

Hongyi Pan, Ziliang Hong, Gorkem Durak et al.

Federated learning (FL) has emerged as a promising paradigm for collaboratively training deep learning models across institutions without exchanging sensitive medical data. However, its effectiveness is often hindered by limited data availability and non-independent, identically distributed data across participating clients, which can degrade model performance and generalization. To address these challenges, we propose a generative AI based data augmentation framework that integrates synthetic image sharing into the federated training process for breast cancer diagnosis via ultrasound images. Specifically, we train two simple class-specific Deep Convolutional Generative Adversarial Networks: one for benign and one for malignant lesions. We then simulate a realistic FL setting using three publicly available breast ultrasound image datasets: BUSI, BUS-BRA, and UDIAT. FedAvg and FedProx are adopted as baseline FL algorithms. Experimental results show that incorporating a suitable number of synthetic images improved the average AUC from 0.9206 to 0.9237 for FedAvg and from 0.9429 to 0.9538 for FedProx. We also note that excessive use of synthetic data reduced performance, underscoring the importance of maintaining a balanced ratio of real and synthetic samples. Our findings highlight the potential of generative AI based data augmentation to enhance FL results in the breast ultrasound image classification task.

Hongyi Pan, Emadeldeen Hamdan, Xin Zhu et al.

Central to the Transformer architectures' effectiveness is the self-attention mechanism, a function that maps queries, keys, and values into a high-dimensional vector space. However, training the attention weights of queries, keys, and values is non-trivial from a state of random initialization. In this paper, we propose two methods. (i) We first address the initialization problem of Vision Transformers by introducing a simple, yet highly innovative, initialization approach utilizing discrete cosine transform (DCT) coefficients. Our proposed DCT-based \textit{attention} initialization marks a significant gain compared to traditional initialization strategies; offering a robust foundation for the attention mechanism. Our experiments reveal that the DCT-based initialization enhances the accuracy of Vision Transformers in classification tasks. (ii) We also recognize that since DCT effectively decorrelates image information in the frequency domain, this decorrelation is useful for compression because it allows the quantization step to discard many of the higher-frequency components. Based on this observation, we propose a novel DCT-based compression technique for the attention function of Vision Transformers. Since high-frequency DCT coefficients usually correspond to noise, we truncate the high-frequency DCT components of the input patches. Our DCT-based compression reduces the size of weight matrices for queries, keys, and values. While maintaining the same level of accuracy, our DCT compressed Swin Transformers obtain a considerable decrease in the computational overhead.

Abhijit Das, Debesh Jha, Vandan Gorade et al.

Computer-aided segmentation methods can assist medical personnel in improving diagnostic outcomes. While recent advancements like UNet and its variants have shown promise, they face a critical challenge: balancing accuracy with computational efficiency. Shallow encoder architectures in UNets often struggle to capture crucial spatial features, leading in inaccurate and sparse segmentation. To address this limitation, we propose a novel \underline{P}rogressive \underline{A}ttention based \underline{M}obile \underline{UNet} (\underline{PAM-UNet}) architecture. The inverted residual (IR) blocks in PAM-UNet help maintain a lightweight framework, while layerwise \textit{Progressive Luong Attention} ($\mathcal{PLA}$) promotes precise segmentation by directing attention toward regions of interest during synthesis. Our approach prioritizes both accuracy and speed, achieving a commendable balance with a mean IoU of 74.65 and a dice score of 82.87, while requiring only 1.32 floating-point operations per second (FLOPS) on the Liver Tumor Segmentation Benchmark (LiTS) 2017 dataset. These results highlight the importance of developing efficient segmentation models to accelerate the adoption of AI in clinical practice.

Ziliang Hong, Halil Ertugrul Aktas, Andrea Mia Bejar et al.

We present the first federated learning (FL) approach for pancreas part(head, body and tail) segmentation in MRI, addressing a critical clinical challenge as a significant innovation. Pancreatic diseases exhibit marked regional heterogeneity cancers predominantly occur in the head region while chronic pancreatitis causes tissue loss in the tail, making accurate segmentation of the organ into head, body, and tail regions essential for precise diagnosis and treatment planning. This segmentation task remains exceptionally challenging in MRI due to variable morphology, poor soft-tissue contrast, and anatomical variations across patients. Our novel contribution tackles two fundamental challenges: first, the technical complexity of pancreas part delineation in MRI, and second the data scarcity problem that has hindered prior approaches. We introduce a privacy-preserving FL framework that enables collaborative model training across seven medical institutions without direct data sharing, leveraging a diverse dataset of 711 T1W and 726 T2W MRI scans. Our key innovations include: (1) a systematic evaluation of three state-of-the-art segmentation architectures (U-Net, Attention U-Net,Swin UNETR) paired with two FL algorithms (FedAvg, FedProx), revealing Attention U-Net with FedAvg as optimal for pancreatic heterogeneity, which was never been done before; (2) a novel anatomically-informed loss function prioritizing region-specific texture contrasts in MRI. Comprehensive evaluation demonstrates that our approach achieves clinically viable performance despite training on distributed, heterogeneous datasets.

Hongyi Pan, Gorkem Durak, Elif Keles et al.

Pancreatic cancer is projected to be the second-deadliest cancer by 2030, making early detection critical. Intraductal papillary mucinous neoplasms (IPMNs), key cancer precursors, present a clinical dilemma, as current guidelines struggle to stratify malignancy risk, leading to unnecessary surgeries or missed diagnoses. Here, we developed Cyst-X, an AI framework for IPMN risk prediction trained on a unique, multi-center dataset of 1,461 MRI scans from 764 patients. Cyst-X achieves significantly higher accuracy (AUC = 0.82) than both the established Kyoto guidelines (AUC = 0.75) and expert radiologists, particularly in correct identification of high-risk lesions. Clinically, this translates to a 20% increase in cancer detection sensitivity (87.8% vs. 64.1%) for high-risk lesions. We demonstrate that this performance is maintained in a federated learning setting, allowing for collaborative model training without compromising patient privacy. To accelerate research in early pancreatic cancer detection, we publicly release the Cyst-X dataset and models, providing the first large-scale, multi-center MRI resource for pancreatic cyst analysis.

Hongyi Pan, Xin Zhu, Salih Atici et al.

In this paper, we propose a novel Hadamard Transform (HT)-based neural network layer for hybrid quantum-classical computing. It implements the regular convolutional layers in the Hadamard transform domain. The idea is based on the HT convolution theorem which states that the dyadic convolution between two vectors is equivalent to the element-wise multiplication of their HT representation. Computing the HT is simply the application of a Hadamard gate to each qubit individually, so the HT computations of our proposed layer can be implemented on a quantum computer. Compared to the regular Conv2D layer, the proposed HT-perceptron layer is computationally more efficient. Compared to a CNN with the same number of trainable parameters and 99.26\% test accuracy, our HT network reaches 99.31\% test accuracy with 57.1\% MACs reduced in the MNIST dataset; and in our ImageNet-1K experiments, our HT-based ResNet-50 exceeds the accuracy of the baseline ResNet-50 by 0.59\% center-crop top-1 accuracy using 11.5\% fewer parameters with 12.6\% fewer MACs.

Hongyi Pan, Diaa Badawi, Ahmet Enis Cetin

Convolution has been the core operation of modern deep neural networks. It is well-known that convolutions can be implemented in the Fourier Transform domain. In this paper, we propose to use binary block Walsh-Hadamard transform (WHT) instead of the Fourier transform. We use WHT-based binary layers to replace some of the regular convolution layers in deep neural networks. We utilize both one-dimensional (1-D) and two-dimensional (2-D) binary WHTs in this paper. In both 1-D and 2-D layers, we compute the binary WHT of the input feature map and denoise the WHT domain coefficients using a nonlinearity which is obtained by combining soft-thresholding with the tanh function. After denoising, we compute the inverse WHT. We use 1D-WHT to replace the $1\times 1$ convolutional layers, and 2D-WHT layers can replace the 3$\times$3 convolution layers and Squeeze-and-Excite layers. 2D-WHT layers with trainable weights can be also inserted before the Global Average Pooling (GAP) layers to assist the dense layers. In this way, we can reduce the number of trainable parameters significantly with a slight decrease in trainable parameters. In this paper, we implement the WHT layers into MobileNet-V2, MobileNet-V3-Large, and ResNet to reduce the number of parameters significantly with negligible accuracy loss. Moreover, according to our speed test, the 2D-FWHT layer runs about 24 times as fast as the regular $3\times 3$ convolution with 19.51\% less RAM usage in an NVIDIA Jetson Nano experiment.

Hongyi Pan, Diaa Badawi, Runxuan Miao et al.

Power iteration is a fundamental algorithm in data analysis. It extracts the eigenvector corresponding to the largest eigenvalue of a given matrix. Applications include ranking algorithms, recommendation systems, principal component analysis (PCA), among many others. In this paper, we introduce multiplication-avoiding power iteration (MAPI), which replaces the standard $\ell_2$-inner products that appear at the regular power iteration (RPI) with multiplication-free vector products which are Mercer-type kernel operations related with the $\ell_1$ norm. Precisely, for an $n\times n$ matrix, MAPI requires $n$ multiplications, while RPI needs $n^2$ multiplications per iteration. Therefore, MAPI provides a significant reduction of the number of multiplication operations, which are known to be costly in terms of energy consumption. We provide applications of MAPI to PCA-based image reconstruction as well as to graph-based ranking algorithms. When compared to RPI, MAPI not only typically converges much faster, but also provides superior performance.

Hongyi Pan, Diaa Badawi, Erdem Koyuncu et al.

We consider a family of vector dot products that can be implemented using sign changes and addition operations only. The dot products are energy-efficient as they avoid the multiplication operation entirely. Moreover, the dot products induce the $\ell_1$-norm, thus providing robustness to impulsive noise. First, we analytically prove that the dot products yield symmetric, positive semi-definite generalized covariance matrices, thus enabling principal component analysis (PCA). Moreover, the generalized covariance matrices can be constructed in an Energy Efficient (EEF) manner due to the multiplication-free property of the underlying vector products. We present image reconstruction examples in which our EEF PCA method result in the highest peak signal-to-noise ratios compared to the ordinary $\ell_2$-PCA and the recursive $\ell_1$-PCA.

Hongyi Pan, Diaa Dabawi, Ahmet Enis Cetin

In this paper, we propose a novel layer based on fast Walsh-Hadamard transform (WHT) and smooth-thresholding to replace $1\times 1$ convolution layers in deep neural networks. In the WHT domain, we denoise the transform domain coefficients using the new smooth-thresholding non-linearity, a smoothed version of the well-known soft-thresholding operator. We also introduce a family of multiplication-free operators from the basic 2$\times$2 Hadamard transform to implement $3\times 3$ depthwise separable convolution layers. Using these two types of layers, we replace the bottleneck layers in MobileNet-V2 to reduce the network's number of parameters with a slight loss in accuracy. For example, by replacing the final third bottleneck layers, we reduce the number of parameters from 2.270M to 540K. This reduces the accuracy from 95.21\% to 92.98\% on the CIFAR-10 dataset. Our approach significantly improves the speed of data processing. The fast Walsh-Hadamard transform has a computational complexity of $O(m\log_2 m)$. As a result, it is computationally more efficient than the $1\times1$ convolution layer. The fast Walsh-Hadamard layer processes a tensor in $\mathbb{R}^{10\times32\times32\times1024}$ about 2 times faster than $1\times1$ convolution layer on NVIDIA Jetson Nano computer board.