Vision-language models (VLMs) have demonstrated substantial promise in medical image segmentation by utilizing radiology reports as prompts to segment regions of interest. However, VLM deployment in clinical settings is challenged by two intertwined issues: i) demographic bias, where performance varies across demographic groups, and ii) prompt sensitivity, where semantically similar prompts yield inconsistent outputs. These challenges are interconnected; demographic underrepresentation can worsen a model’s sensitivity to prompts, and prompt instability can more heavily affect certain demographic groups. In this study, we present FairVLM, a unified framework that addresses both demographic disparity and prompt sensitivity in VLMs. FairVLM integrates three key components: (1) Semantic-Retaining Counterfactual Prompting (SRCP), which generates clinically consistent and diverse prompt variations via large language models; (2) Demographic-Aware Feature Normalization (DAFN), a lightweight module that mitigates latent representation bias across demographic groups; and (3) a Fairness-Calibrated Loss (FCL) that explicitly penalizes performance disparities while encouraging prompt consistency. Extensive evaluations on the Harvard-FairSeg dataset show that FairVLM significantly improves equity-scaled segmentation. It also reduces demographic disparity (DI) by over 65% and relative performance gap (RPG) by over 60%, while maintaining or boosting overall accuracy. FairVLM is robust to prompt changes, with less than 0.5% performance drop across varied prompts, and also generalizes well on unseen datasets. These findings present FairVLM as a new state-of-the-art, robust, and adaptable framework for a fair, prompt-invariant vision-language model. Code and data are available at https://github.com/Rahman-Motiur/FairVLM.

In this paper, we present a wireless closed-loop neuromodulation platform dedicated to diabetes management through implants in various body locations, such as the brain, stomach, and pancreas. The system's key components comprise a System-on-Chip (SoC) featuring a maximum 64-channel neural recording block to explore both neural roots and muscles involved in diabetes emergence and interactions within diverse organs with superior spatial resolution, an energy-efficient Impulse Radio Ultra-Wideband (IR-UWB) wireless transmitter, and an 8-channel stimulator for highly selective organ targeting. These capabilities are further enhanced by an edge-based Artificial Intelligence (AI) mechanism for real-time analysis. Simulation results show a total power consumption of 31.16 μW per channel for the recording and wireless transmission units, while the stimulator, powered by an inductive link, offers adjustable output with a maximum current of 3 mA and a voltage compliance of 9 V. Additionally, AI implemented as a four-layer neural network (NN) model using a field-programmable gate array (FPGA) on a wearable system achieved an accuracy of 98.06% in test sets.

Precise medical image segmentation is important for automating diagnosis and treatment planning in healthcare. While images present the most significant information for segmenting organs using deep learning models, text reports also provide complementary details that can be leveraged to improve segmentation precision. Performance improvement depends on the proper utilization of text reports and the corresponding images. Most attention modules focus on single-modality computation of spatial, channel, or pixel-level attention. They are ineffective in cross-modal alignment, raising issues in multi-modal scenarios. This study addresses these gaps by presenting a text-assisted vision (TAV) model for medical image segmentation with a novel attention computation module named tri-guided attention module (TGAM). TGAM computes visual-visual, language-language, and language-visual attention, enabling the model to understand the important features and correlation between images and medical notes. This module helps the model identify the relevant features within images, text annotations, and text annotations to visual interactions. We incorporate an attention gate (AG) that modulates the influence of TGAM, ensuring it does not overflow the encoded features with irrelevant or redundant information, while maintaining their uniqueness. We evaluated the performance of TAV on two popular datasets containing images and corresponding text annotations. We find TAV to be a new state-of-the-art model, as it improves the performance by 2-7% compared to other models. Extensive experiments were performed to demonstrate the effectiveness of each component of the proposed model.

One of the common and promising deep learning approaches used for medical image segmentation is transformers, as they can capture long-range dependencies among the pixels by utilizing self-attention. Despite being successful in medical image segmentation, transformers face limitations in capturing local contexts of pixels in multimodal dimensions. We propose a Medical Image Segmentation Transformer (MIST) incorporating a novel Convolutional Attention Mixing (CAM) decoder to address this issue. MIST has two parts- a pre-trained multi-axis vision transformer (MaxViT) is used as an encoder, and the encoded feature representation is passed through the CAM decoder for segmenting the images. In the CAM decoder, an attention-mixer combining multi-head self-attention, spatial attention, and squeeze and excitation attention modules is introduced to capture long-range dependencies in all spatial dimensions. Moreover, to enhance spatial information gain, deep and shallow convolutions are used for feature extraction and receptive field expansion, respectively. The integration of low-level and high-level features from different network stages is enabled by skip connection, allowing MIST to suppress unnecessary information. The experiments show that our MIST transformer with CAM decoder outperforms the state-of-the-art models specifically designed for medical image segmentation on the ACDC and Synapse datasets. Our results also demonstrate that adding the CAM decoder with a hierarchical transformer improves the segmentation performance significantly. Our model with data and code is publicly available on GitHub1.

Designing a machine learning based network intrusion detection system (IDS) with high-dimensional features can lead to prolonged classification processes. This is while low-dimensional features can reduce these processes. Moreover, classification of network traffic with imbalanced class distributions has posed a significant drawback on the performance attainable by most well-known classifiers. With the presence of imbalanced data, the known metrics may fail to provide adequate information about the performance of the classifier. This study first uses Principal Component Analysis (PCA) as a feature dimensionality reduction approach. The resulting low-dimensional features are then used to build various classifiers such as Random Forest (RF), Bayesian Network, Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA) for designing an IDS. The experimental findings with low-dimensional features in binary and multi-class classification show better performance in terms of Detection Rate (DR), F-Measure, False Alarm Rate (FAR), and Accuracy. Furthermore, in this paper, we apply a Multi-Class Combined performance metric Combi ned Mc with respect to class distribution through incorporating FAR, DR, Accuracy, and class distribution parameters. In addition, we developed a uniform distribution based balancing approach to handle the imbalanced distribution of the minority class instances in the CICIDS2017 network intrusion dataset. We were able to reduce the CICIDS2017 dataset's feature dimensions from 81 to 10 using PCA, while maintaining a high accuracy of 99.6% in multi-class and binary classification.

Machine learning algorithms are effective means applied to wireless intrusion detection systems (WIDS) in an attempt to protect computing resources against unauthorized access. A key aspect for an improved WIDS based on machine learning classification is features selection. This paper considers multiclass classification that utilizes four effective features sets of 32, 10, 7 and 5 features, respectively. The classes represent 15 types of 802.11 MAC layer attacks, and the features represent 802.11 frame fields information. The experimental results utilized the Aegean Wi-Fi Intrusion Dataset (AWID) to evaluate the performance of seven well-known machine learning classifiers, namely, AdaBoost, Random Forest, Random Tree, J48, logit Boost, Multi-Layer Perceptron, and ZeroR with respect to the selected features set. The presented work outperforms previous related work in terms of number of classes, features and accuracy. The proposed system using the Random Forest algorithm and 32 features achieves a maximum accuracy of 99.64%. By using logit Boost with five features, we achieved a maximum accuracy of 99.53%.

In cybersecurity, machine learning approaches can predict and detect threats before they result in major security incidents. The design and performance of an effective machine learning based Intrusion Detection System (IDS) depends upon the selected attributes and the classifier model. This paper considers multi-class classification for the Aegean Wi-Fi Intrusion Dataset (AWID) where classes represent 17 types of the IEEE 802.11 MAC Layer attacks. The proposed work extracts four attribute sets of 32, 10, 7 and 5 attributes, respectfully. The classifiers achieved high accuracy with minimum false positive rates, and the presented work outperforms previous related work in terms of number of classes, attributes and accuracy. The proposed work achieved maximum accuracy of 99.64% for Random Forest with supply test and 99.99% using the 10-fold cross validation approach for Random Forest and J48.

Deep brain stimulation (DBS) has been widely practiced for the treatment of advanced Parkinsons disease (PD) but the underlying mechanism is still not well understood. Subthalamic Nucleus (STN), Globus Pallidus externa (GPe) and Globus Pallidus interna (GPi) neurons are the common DBS target sites, which are selected based on the patients symptoms. The DBS pulse shape must also guarantee activation of the desired neurons and optimized energy consumption. In this paper, we apply energy efficient Gaussian DBS signals on different targets in a computational model of the basal ganglia. The results shows that Gaussian signals outperform the widely-used rectangular signals in terms of desynchronization of the GPi neurons and the total energy consumed by the DBS process. Our quantitative results suggest that targeting STN neurons for DBS (STN-DBS) is beneficial for PD patients with axial and cardinal symptoms, while dyskinesia is better treated by GPi-DBS, and improving bradykinesia and akinesia symptoms of PD is mostly achieved by GPe-DBS.

This paper presents a conceptual framework of a virtual reality therapy to assist individuals, especially lung cancer patients or those with breathing disorders to regulate their breath through real-time analysis of respiration movements using a smartphone. Virtual reality technology is an attractive means for medical simulations and treatment, particularly for patients with cancer. The theories, methodologies and approaches, and real-world dynamic contents for all the components of this virtual reality therapy (VRT) via a conceptual framework using the smartphone will be discussed. The architecture and technical aspects of the offshore platform of the virtual environment will also be presented.

Monitoring breath and identifying breathing movements have settled importance in many biomedical research areas, especially in the treatment of those with breathing disorders, e.g., lung cancer patients. Moreover, virtual reality (VR) revolution and their implementations on ubiquitous hand-held devices have a lot of implications, which could be used as a simulation technology for healing purposes. In this paper, a novel method is proposed to detect and classify breathing movements. The overall VR framework is intended to encourage the subjects regulate their breath by classifying the breathing movements in real time. This paper focuses on a portion of the overall VR framework that deals with classifying the acoustic signal of respiration movements. We employ Mel-frequency cepstral coefficients (MFCCs) along with speech segmentation techniques using voice activity detection and linear thresholding to the acoustic signal of breath captured using a microphone to depict the differences between inhale and exhale in frequency domain. For every subject, 13 MFCCs of all voiced segments are computed and plotted. The inhale and exhale phases are differentiated using the sixth MFCC order, which carries important classification information. Experimental results on a number of individuals verify our proposed classification methodology.

Breathing disorders are generally associated with the lung cancer disease. Through daily treatment, lung cancer patients often use a traditional spirometer to measure their lung capacity. However, the use of a spirometer device for accurate measurement requires some sort of training and adjustment, which may be inconvenient for certain groups of patients, especially the elderly. In addition, the spirometer readings can become unreliable if the measurements are not taken as instructed. On the other hand, a microphone, say the microphone on a hand-held device such as a smart-phone, can easily capture the acoustic signal of breath without certain instructions. This signal can then be processed to estimate the lung capacity. In this paper, we propose a methodology through the simply recorded acoustic signal of breath, splitting the breathing cycle to inhale, pause, exhale and pause phases to measure the depth of the breath along with the time duration and signal energy of the breathing phases. We show how these computed parameters are used to estimate the lung size with a high degree of accuracy. This work is part of a virtual reality platform embedded within a smart-phone to assist lung cancer patients regulate their breath. Furthermore, the lung capacity estimation methodology proposed in this paper can also be used to aid patients with other breathing disorders.

Pressure ulcer is a critical problem for bed-ridden and wheelchair-bound patients, diabetics, and the elderly. Patients need to be regularly repositioned to prevent excessive pressure on a single area of body, which can lead to ulcers. Pressure ulcers are extremely costly to treat and may lead to several other health problems, including death. The current standard for prevention is to reposition at-risk patients every two hours. Even if it is done properly, a fixed schedule is not sufficient to prevent all ulcers. Moreover, it may result in nurses being overworked by turning some patients too frequently. In this paper, we present an algorithm for finding a nurse-effort optimal repositioning schedule that prevents pressure ulcer formation for a finite planning horizon. Our proposed algorithm uses data from a commercial pressure mat assembled on the beds surface and provides a sequence of next positions and the time of repositioning for each patient.

Pressure ulcer is a severe threat for immobilized and peripheral neuropathic patients such as bed-ridden, elderly, and diabetics. Once developed, the complication of pressure ulcer causes pain, suffering, and longer hospitalization for the patients. Additionally, pressure ulcer management imposes a serious burden on the health care providers. The optimal strategy to deal with pressure ulcers is prevention. The current standard for prevention is to reposition at-risk patients every two hours. But, each patient has different needs based on overall vulnerability and damaged skin areas. A fixed schedule may either result in some patients getting ulcers, or nurses being overworked by turning some patients too frequently. In this paper, we present an efficient algorithm to find a repositioning schedule for bed- bound patients based on their risk of ulcer development. Our proposed algorithm uses data from a commercial pressure mat assembled on the bed's surface and provides a sequence of next positions and the time of repositioning for each patient. Our patient-specific turning schedule minimizes the overall cost of nursing staff involvement in repositioning the patients while simultaneously decreases the chance of pressure ulcer formation.

Network intrusion detection systems continuously monitor the network traffic in order to identify any traces of suspicious activities such as worm, viruses or spam. One attractive technique for identifying potential Internet threats is detecting previously unknown, but common sub-strings that appear very frequently in data packets. In this paper, we propose a novel architectural platform that thoroughly analyzes the network traffic behavior in terms of repetitions to identify potential Internet threats. The main idea is to use a two-phase hashing system and small memory units functioning in parallel to achieve a high-throughput and memory efficient behavioral analysis engine. The system performs behavioral analysis on selected information/user(s) and builds a bell-shaped curve for normal traffic using parallel counters. Our traffic behavioral analysis system has been fully prototyped on Altera Stratix FPGA. Experimental results verify that our system can support line speed of gigabit-rates with very negligible false positive and negative rates.

Recent trends in clinical and telemedicine applications highly demand automation in (electrocardiogram) ECG signal processing and heart beat classification. A real-time patient-adaptive cardiac profiling scheme using repetition detection is proposed in this paper. We introduce a novel local ECG beat classifier to profile each patient's normal cardiac behavior. As ECG morphologies vary from person to person, and even for each person, it can vary depending on the person's physical condition, having such profile is essential for various diagnosis (e.g. arrhythmia) purposes, and can successfully raise an early warning flag for the abnormal cardiac behavior of any individual. Experimental results show that our technique follows the MIT/BIH arrhythmia database annotations with high accuracy.

Recent trends in clinical and telemedicine applications highly demand automation in electrocardiogram (ECG) signal processing and heart beat classification. A patient-adaptive cardiac profiling scheme using repetition-detection concept is proposed in this paper. We first employ an efficient wavelet-based beat-detection mechanism to extract precise fiducial ECG points. Then, we implement a novel local ECG beat classifier to profile each patient's normal cardiac behavior. ECG morphologies vary from person to person and even for each person, it can vary over time depending on the person's physical condition and/or environment. Having such profile is essential for various diagnosis (e.g., arrhythmia) purposes. One application of such profiling scheme is to automatically raise an early warning flag for the abnormal cardiac behavior of any individual. Our extensive experimental results on the MIT-BIH arrhythmia database show that our technique can detect the beats with 99.59% accuracy and can identify abnormalities with a high classification accuracy of 97.42%.

Most conventional packet classifiers find only the highest priority filter that matches the arriving packet. However, new networking applications such as network intrusion detection systems and load balancers require all (or the first few) matching packets during classification. In this paper, two TCAM-based architectures for multi-match search are introduced. The first one is a renovated TCAM design that can find all or the first r matches in a packet filter set. The second architecture is a novel partitioning scheme based on filter intersection properties allowing us to use off-the-shelf TCAMs for multi-match packet classification. Our classifier engine finds all matches in exactly one conventional TCAM cycle while reducing the power consumption by at least two orders of magnitude, which is far better than the existing hardware based designs.

Network Intrusion Detection Systems (NIDS) monitor network traffic to detect attacks or unauthorized activities. Traditional NIDSes search for patterns that match typical network compromise or remote hacking attempts. However, newer networking applications require finding the frequently repeated strings in a packet stream for further investigation of potential attack attempts. Finding frequently repeated strings within a given time frame of the packet stream has been quite efficient to detect polymorphic worm outbreaks. A novel real-time worm outbreak detection system using two-phase hashing and monitoring repeated common substrings is proposed in this article. We use the concept of shared counters to minimize the memory cost while efficiently sifting through suspicious strings. The worm outbreak system has been prototyped on Altera Stratix FPGA. We have tested the system for various settings and packet stream sizes. Experimental results verify that our system can support line speed of gigabit-rates with negligible false positive and negative rates.

This paper introduces a new basic building block based on Non-deterministic Finite Automata (NFA) hardware implementation to support complex constraint repetitions in regular expressions. This block is a customized counter capable of handling any type of constraint repetition, applicable to any sub-regular expression. We also apply optimization techniques to reduce the area and improve the overall performance.

Recent network intrusion detection systems (NIDS) use regular expressions to represent suspicious or malicious character sequences in packet payloads in a more efficient way. This paper introduces a new basic building block based on Non- deterministic Finite Automata (NFA) hardware implementation to support complex constraint repetitions in regular expressions. This block is a customized counter capable of handling any type of constraint repetition, applicable to any sub-regular expression. We also introduce optimization techniques to reduce the area and improve the overall performance. We have examined SNORT IDS regular expressions by taking advantage of the basic NFA building blocks, our proposed counting block and our proposed optimization techniques. We report experimental results for our architecture that verify area saving and performance improvement.

Most conventional packet classifiers find the highest priority filter that matches the packet. However, new networking applications such as network intrusion detection systems (NIDS) and load balancers require all (or the first few) matching results in packet classification. A TCAM -based architecture optimized for multiple match search is introduced in this paper. We propose a renovated TCAM design that can find all or the first r matches in a packet filter set. Our scheme partitions the filter set and performs the multi-match search on one partition only. A VLSI implementation of our classifier in 0.18 mum technology can achieve speed that is 1-2 order of magnitude higher than software based approaches. Power consumption is reduced by 40% which is far better than the conventional hardware based multi-match designs.