A Comparative Study and Systematic Analysis of XAI Models and their Applications in Healthcare

Archives of Computational Methods in Engineering, Journal Year: 2024, Volume and Issue: unknown

Published: April 16, 2024

Language: Английский

---

A review of evaluation approaches for explainable AI with applications in cardiology

Artificial Intelligence Review, Journal Year: 2024, Volume and Issue: 57(9)

Published: Aug. 9, 2024

Abstract Explainable artificial intelligence (XAI) elucidates the decision-making process of complex AI models and is important in building trust model predictions. XAI explanations themselves require evaluation as to accuracy reasonableness context use underlying model. This review details cardiac applications has found that, studies examined, 37% evaluated quality using literature results, 11% used clinicians domain-experts, proxies or statistical analysis, with remaining 43% not assessing at all. We aim inspire additional within healthcare, urging researchers only apply methods but systematically assess resulting explanations, a step towards developing trustworthy safe models.

Language: Английский

---

Multi-modality image fusion for medical assistive technology management based on hybrid domain filtering

Expert Systems with Applications, Journal Year: 2022, Volume and Issue: 209, P. 118283 - 118283

Published: July 28, 2022

Language: Английский

---

Paving the Roadmap for XAI and IML in Healthcare: Data-Driven Discoveries and the FIXAIH Framework

Published: Jan. 1, 2025

Integrating Explainable Artificial Intelligence (XAI) and Interpretable Machine Learning (IML) in healthcare enhances trust transparency, crucial for outcomes that directly affect patient care. In this paper, we design a machine learning-based analysis tool to systematically analyze dataset of 5,083 academic articles, focusing on how XAI IML can be effectively integrated into healthcare. Our identifies categorizes 13 key parameters across three macro-parameters: Research Methods, Health Disorders, Disease Prevention. This categorization, informed by focused review over 200 helped clarify specific applications challenges associated with settings. These illustrate the profound impact advancing healthcare, from improving diagnostic accuracy treatment efficacy predicting preventing health risks. Methods enhance analytic capabilities clinical decision-making, Disorders apply managing diseases such as cancer chronic conditions, Prevention uses predictive analytics improve preventive strategies. Based these findings, propose FIXAIH framework, designed operationalize insights actionable guidelines interpretability, explainability, accountability AI systems By offering structured comprehensive guidelines, framework ensures tools are not only technically proficient but also ethically sound easily understandable professionals. paper aims bridge technical-proficiency gap promote practical application technologies, fostering more reliable user-centric approach medical field.

Language: Английский

---

Detection of Atrial Fibrillation in Holter ECG Recordings by ECHOView Images: A Deep Transfer Learning Study

Diagnostics, Journal Year: 2025, Volume and Issue: 15(7), P. 865 - 865

Published: March 28, 2025

Background/Objectives: The timely and accurate detection of atrial fibrillation (AF) is critical from a clinical perspective. Detecting short or transient AF events challenging in 24-72 h Holter ECG recordings, especially when symptoms are infrequent. This study aims to explore the potential deep transfer learning with ImageNet neural networks (DNNs) improve interpretation short-term ECHOView images for presence AF. Methods: Thirty-second images, composed stacked heartbeat amplitudes, were rescaled fit input 18 pretrained DNNs top layers modified binary classification (AF, non-AF). Transfer provided both retrained by training only (513-2048 trainable parameters) fine-tuned slowly (0.38-23.48 M parameters). Results: used 13,536 6624 validation samples two leads IRIDIA-AF database, evenly split between non-AF cases. top-ranked evaluated on 11,400 test independent records EfficientNetV2B1 (96.3% accuracy minimal inter-patient (1%) inter-lead (0.3%) drops), DenseNet-121, -169, -201 (97.2-97.6% (1.4-1.6%) (0.5-1.2%) drops). These models can process shorter episodes tolerable drop up 0.6% 20 s 4-15% 10 s. Case studies present GradCAM heatmaps overlaid raw illustrate model interpretability. Conclusions: In an extended study, we validate that applied through retraining fine-tuning significantly enhance automated diagnoses. provide meaningful interpretability, highlighting regions interest aligned cardiologist focus.

Language: Английский

---

A multi-label deep residual shrinkage network for high-density surface electromyography decomposition in real-time

Journal of NeuroEngineering and Rehabilitation, Journal Year: 2025, Volume and Issue: 22(1)

Published: May 8, 2025

The swift and accurate identification of motor unit spike trains (MUSTs) from surface electromyography (sEMG) is essential for enabling real-time control in neural interfaces. However, the existing sEMG decomposition methods, including blind source separation (BSS) deep learning, have not yet achieved satisfactory performance, due to high latency or low accuracy. This study introduces a novel high-density (HD-sEMG) algorithm named ML-DRSNet, which combines multi-label learning with residual shrinkage network (DRSNet) improve accuracy reduce latency. ML-DRSNet was evaluated on public dataset corresponding MUSTs extracted via convolutional BSS algorithm. An improved (ML-DCNN) also compared against conventional multi-task DCNN (MT-DCNN). These networks were trained tested various window sizes step sizes. With shortest size (20 data points) (10 points), significantly outperformed both ML-DCNN (0.86 ± 0.18 vs. 0.71 0.24, P < 0.001) MT-DCNN 0.66 0.16, precision. Moreover, demonstrated notably lower (15.15 ms) (69.36 (76.96 ms), reduced relative BSS-based methods. proposed algorithms substantially enhance performance decomposing MUSTs, establishing technical foundation neuro-information-driven intention recognition disease assessment.

Language: Английский

---

Intelligent Support for Cardiovascular Diagnosis

Advances in media, entertainment and the arts (AMEA) book series, Journal Year: 2024, Volume and Issue: unknown, P. 64 - 76

Published: Jan. 10, 2024

The AI-CDSS is a powerful tool designed to assist healthcare professionals in making informed and evidence-based decisions patient care. It leverages artificial intelligence algorithms data analysis techniques provide personalized recommendations insights. This system explores the features benefits of AI-CDSS, including analysis, diagnostics treatment recommendations, drug interaction adverse event detection, predictive analytics, real-time monitoring alerts, continuous learning improvement. model also discusses applications AI-driven decision-making systems healthcare, focusing on areas such as cancer diagnosis treatment, chronic disease management, medication optimization, surgical decision support, infectious outbreak radiology medical imaging mental health clinical trials research. Additionally, chapter highlights existing methodologies, deep models like CNNs RNNs, that have shown potential cardiovascular prediction.

Language: Английский

---

Data-Driven Energy Management of an Electric Vehicle Charging Station Using Deep Reinforcement Learning

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 65956 - 65966

Published: Jan. 1, 2024

A charging station that integrates renewable energy sources is a promising solution to address the increasing demand for electric vehicle (EV) without expanding distribution network. An efficient and flexible management strategy essential effectively integrating various EVs. This research work aims develop an Energy Management System (EMS) EV (EVCS) minimizes operating cost of EVCS operator while meeting demands connected The proposed approach employs model-free method leveraging Deep Reinforcement Learning (DRL) identify optimal schedules EVs in real time. Markov Decision Process (MDP) model constructed from perspective operator. real-world scenarios are formulated by considering stochastic nature commuting behavior Various DRL algorithms addressing MDPs examined, their performances empirically compared. Notably, Truncated Quantile Critics (TQC) algorithm emerges as superior choice, yielding enhanced performance. simulation findings show EMS can offer control strategy, reducing operators compared other benchmark methods.

Language: Английский

---

Enhancing Fake Image Detection: A Novel Two-Step Approach Combining GANs and CNNs

Procedia Computer Science, Journal Year: 2024, Volume and Issue: 235, P. 810 - 819

Published: Jan. 1, 2024

The proliferation of fake images in today's digital landscape poses a significant threat to various domains, including media integrity, social media, and online security. Recognizing the urgent need distinguish real from their deceptive counterparts, this paper underscores importance developing robust detection system. While substantial efforts have been made realms computer vision deep learning, advent Generative Adversarial Networks (GANs) has added new layer complexity challenge. In response these evolving threats, we present novel two-step methodology for detecting images, with specific focus on those generated by GANs. Our approach harnesses combined strengths GANs traditional Convolutional Neural (CNNs), offering comprehensive solution that significantly enhances accuracy identifying both machine-generated images. results our experiments demonstrate efficacy methodology. Using CNNs alone, achieved training 87%. However, when employing collaborative power CNNs, model exhibited remarkable rate 94.4%. This improvement superiority GANs+CNN approach, suggesting its potential as groundbreaking realm image detection. research opens up horizons fields such forensics, monitoring, security, where ability discern genuine content manipulated or synthetic is paramount importance. promising outcomes study not only provide an immediate effective but also pave way further exploration innovation critical area

Language: Английский

---

Interpretable Machine Learning Model for Breast Cancer Prediction Using LIME and SHAP

2022 IEEE 7th International conference for Convergence in Technology (I2CT), Journal Year: 2024, Volume and Issue: unknown

Published: April 5, 2024

Breast cancer, which occurs in both men and women, causes approximately 10 lakh deaths globally has no specific risk factors. The time frame of the treatment is a long-drawn process based on person, type its level spread. It imperative to detect this cancer early order prevent mortality. Given prediction's significance, an accurate breast prediction model must be developed. This study explores Cancer Prediction dataset, applies SMOTE (Synthetic Minority Over-sampling Technique) balance proposes effective Machine Learning (ML) fused with Explainable AI provide health professionals explanations. ML algorithms are analyzed before after applying Principal Component Analysis (PCA), visualization performed using t-SNE. algorithms, Support Vector (SVM), k-Nearest Neighbor (kNN), Random Forest (RF), Stochastic gradient Descent, XGBoost, Gradient Boosting, Decision Tree (DT), Naïve Bayes trained dataset. seen that RF outperforms other models considered 95.9% accuracy. To understand weightage features by best trust doctors, (XAI) packages LIME (Local Interpretable Model-agnostic Explanations), SHAP (SHapley Additive exPlanations) used. XAI techniques empowers clinicians actionable insights for more informed diagnosis decision-making.

Language: Английский

---