Explainable artificial intelligence approaches for COVID-19 prognosis prediction using clinical markers

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Jan. 20, 2024

Language: Английский

---

Marine Vessel Classification and Multivariate Trajectories Forecasting Using Metaheuristics-Optimized eXtreme Gradient Boosting and Recurrent Neural Networks

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(16), P. 9181 - 9181

Published: Aug. 11, 2023

Maritime vessels provide a wealth of data concerning location, trajectories, and speed. However, while these are meticulously monitored logged to maintain course, they can also meta information. This work explored the potential data-driven techniques applied artificial intelligence (AI) tackle two challenges. First, vessel classification was through use extreme gradient boosting (XGboost). Second, trajectory time series forecasting tackled long-short-term memory (LSTM) networks. Finally, due strong dependence AI model performance on proper hyperparameter selection, boosted version well-known particle swarm optimization (PSO) algorithm introduced specifically for tuning hyperparameters models used in this study. The methodology real-world automatic identification system (AIS) both marine forecasting. Boosted PSO (BPSO) compared contemporary optimizers showed promising outcomes. XGBoost tuned using attained an overall accuracy 99.72% problem, LSTM mean square error (MSE) 0.000098 prediction challenge. A rigid statistical analysis performed validate outcomes, explainable principles were determined best-performing models, gain better understanding feature impacts decisions.

Language: Английский

---

Detection of renal cell hydronephrosis in ultrasound kidney images: a study on the efficacy of deep convolutional neural networks

PeerJ Computer Science, Journal Year: 2024, Volume and Issue: 10, P. e1797 - e1797

Published: Jan. 23, 2024

In the realm of medical imaging, early detection kidney issues, particularly renal cell hydronephrosis, holds immense importance. Traditionally, identification such conditions within ultrasound images has relied on manual analysis, a labor-intensive and error-prone process. However, in recent years, emergence deep learning-based algorithms paved way for automation this domain. This study aims to harness power learning models autonomously detect hydronephrosis taken close proximity kidneys. State-of-the-art architectures, including VGG16, ResNet50, InceptionV3, innovative Novel DCNN, were put test subjected rigorous comparisons. The performance each model was meticulously evaluated, employing metrics as F1 score, accuracy, precision, recall. results paint compelling picture. DCNN outshines its peers, boasting an impressive accuracy rate 99.8%. same arena, InceptionV3 achieved notable 90% ResNet50 secured 89%, VGG16 reached 85%. These outcomes underscore DCNN's prowess images. Moreover, offers detailed view model's through confusion matrices, shedding light their abilities categorize true positives, negatives, false negatives. regard, exhibits remarkable proficiency, minimizing both positives conclusion, research underscores supremacy automating With exceptional minimal error rates, stands promising tool healthcare professionals, facilitating early-stage diagnosis treatment. Furthermore, convergence hold potential enhancement further exploration, testing larger more diverse datasets investigating optimization strategies.

Language: Английский

---

Empowering Glioma Prognosis With Transparent Machine Learning and Interpretative Insights Using Explainable AI

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 31697 - 31718

Published: Jan. 1, 2024

The primary objective of this research is to create a reliable technique determine whether patient has glioma, specific kind brain tumour, by examining various diagnostic markers, using variety machine learning as well deep approaches, and involving XAI (explainable artificial intelligence) methods. Through the integration data, including medical records, genetic profiles, algorithms have ability predict how each individual will react different interventions. To guarantee regulatory compliance inspire confidence in AI-driven healthcare solutions, incorporated. Machine methods employed study includes Random Forest, decision trees, logistic regression, KNN, Adaboost, SVM, Catboost, LGBM classifier, Xgboost whereas include ANN CNN. Four alternative strategies, SHAP, Eli5, LIME, QLattice algorithm, are comprehend predictions model. Xgboost, ML model achieved accuracy, precision, recall, f1 score, AUC 88%, 82%, 94%, 92%, respectively. best characteristics according techniques IDH1, Age at diagnosis, PIK3CA, ATRX, PTEN, CIC, EGFR TP53. By applying data analytic techniques, provide professionals with practical tool that enhances their capacity for decision-making, resource management, ultimately raises bar care. Medical experts can customise treatments improve outcomes taking into account patient's particular characteristics. provides justifications foster faith amongst patients who must rely on AI-assisted diagnosis treatment recommendations.

Language: Английский

---

Artificial Intelligence-Based Classification of CT Images Using a Hybrid SpinalZFNet

Interdisciplinary Sciences Computational Life Sciences, Journal Year: 2024, Volume and Issue: 16(4), P. 907 - 925

Published: Aug. 21, 2024

Abstract The kidney is an abdominal organ in the human body that supports filtering excess water and waste from blood. Kidney diseases generally occur due to changes certain supplements, medical conditions, obesity, diet, which causes function ultimately leads complications such as chronic disease, failure, other renal disorders. Combining patient metadata with computed tomography (CT) images essential accurately timely diagnosing complications. Deep Neural Networks (DNNs) have transformed fields by providing high accuracy complex tasks. However, computational cost of these models a significant challenge, particularly real-time applications. This paper proposed SpinalZFNet, hybrid deep learning approach integrates architectural strengths Spinal Network (SpinalNet) feature extraction capabilities Zeiler Fergus (ZFNet) classify disease using CT images. unique combination enhanced analysis, significantly improving classification while reducing overhead. At first, acquired are pre-processed median filter, image segmented Efficient (ENet). Later, augmented, different features extracted augmented finally into normal, tumor, cyst, stone SpinalZFNet model. outperformed models, 99.9% sensitivity, 99.5% specificity, precision 99.6%, 99.8% accuracy, 99.7% F1-Score classifying disease. Graphical

Language: Английский

---

Multiple Explainable Approaches to Predict the Risk of Stroke Using Artificial Intelligence

Information, Journal Year: 2023, Volume and Issue: 14(8), P. 435 - 435

Published: Aug. 1, 2023

Stroke occurs when a brain’s blood artery ruptures or the supply is interrupted. Due to rupture obstruction, tissues cannot receive enough and oxygen. common cause of mortality among older people. Hence, loss life severe brain damage can be avoided if stroke recognized diagnosed early. Healthcare professionals discover solutions more quickly accurately using artificial intelligence (AI) machine learning (ML). As result, we have shown how predict in patients heterogeneous classifiers explainable (XAI). The multistack ML models surpassed all other classifiers, with accuracy, recall, precision 96%, respectively. Explainable collection frameworks tools that aid understanding interpreting predictions provided by algorithms. Five diverse XAI methods, such as Shapley Additive Values (SHAP), ELI5, QLattice, Local Interpretable Model-agnostic Explanations (LIME) Anchor, been used decipher model predictions. This research aims enable healthcare provide personalized efficient care, while also providing screening architecture automated revolutionize prevention treatment.

Language: Английский

---

Interpretable Cervical Cell Classification: A Comparative Analysis

Published: Feb. 21, 2024

Cervical cancer is a significant global health issue, and traditional screening methods like Pap smears are labor-intensive may miss some cases. Automation needed, but it faces challenges in terms of interpretability data availability. To address this, the paper proposes using Explainable Artificial Intelligence (XAI) techniques GradCAM, GradCAM++, LRP to improve transparency cervical cell classification model, making novel contribution enhancing trustworthiness automated detection. Using Herlev Dataset, we employ pre-processing, augmentation develop binary achieving 91.94% accuracy with VGG16. The qualitative analysis XAI confirmed that model relied on nucleus cytoplasm features, key indicators malignancy. least mean image entropy 2.4849 steep prediction confidence drop perturbations quantitatively proved Layer-wise Relevance Propagation (LRP) be most effective technique for classification.

Language: Английский

---

A novel NEMONET framework for enhanced RCC detection and staging in CT images

Deleted Journal, Journal Year: 2025, Volume and Issue: 28(1)

Published: Jan. 15, 2025

This study introduces NemoNet, a novel deep-learning framework designed for the automated detection and staging of Renal Cell Carcinoma (RCC) in 3D CT images. Leveraging comprehensive HubMAP RCC dataset, NemoNet integrates encoder-decoder architecture with advanced radiomic feature analysis to enhance tumour segmentation accuracy. The model employs multi-objective loss function balance precision prediction, outperforming traditional architectures like U-Net ResNet. Evaluation metrics, including Dice Coefficient, sensitivity, specificity, indicate superior performance, achieving an accuracy 92% score 0.88. While demonstrates robust results, challenges remain handling variability imaging quality full interpretability. findings suggest that offers significant advancements staging, potential applications personalized oncology treatment planning.

Language: Английский

---

Dynamic Surgical Prioritization: A Machine Learning and XAI-Based Strategy

Technologies, Journal Year: 2025, Volume and Issue: 13(2), P. 72 - 72

Published: Feb. 8, 2025

Surgical waiting lists present significant challenges to healthcare systems, particularly in resource-constrained settings where equitable prioritization and efficient resource allocation are critical. We aim address these issues by developing a novel, dynamic, interpretable framework for prioritizing surgical patients. Our methodology integrates machine learning (ML), stochastic simulations, explainable AI (XAI) capture the temporal evolution of dynamic scores, qp(t), while ensuring transparency decision making. Specifically, we employ Light Gradient Boosting Machine (LightGBM) predictive modeling, simulations account variables competitive interactions, SHapley Additive Explanations (SHAPs) interpret model outputs at both global patient-specific levels. hybrid approach demonstrates strong performance using dataset 205 patients from an otorhinolaryngology (ENT) unit high-complexity hospital Chile. The LightGBM achieved mean squared error (MSE) 0.00018 coefficient determination (R2) value 0.96282, underscoring its high accuracy estimating qp(t). Stochastic effectively captured changes, illustrating that Patient 1’s qp(t) increased 0.50 (at t=0) 1.026 t=10) due growth such as severity urgency. SHAP analyses identified (Sever) most influential variable, contributing substantially non-clinical factors, capacity participate family activities (Lfam), exerted moderating influence. Additionally, our achieves reduction times up 26%, demonstrating effectiveness optimizing prioritization. Finally, strategy combines adaptability interpretability, transparent aligns with evolving patient needs constraints.

Language: Английский

---

Kidney Stone Detection Using Machine Learning Approaches

Advances in computational intelligence and robotics book series, Journal Year: 2025, Volume and Issue: unknown, P. 103 - 136

Published: Jan. 10, 2025

Kidney stones, formed from urine-derived molecules like uric acid and calcium oxalate, are a growing global health issue, affecting 11% of men 9% women. Diseases high blood pressure, diabetes, obesity increase the risk. This study aims to detect kidney stones in CT images using Machine Learning (ML) Deep (DL) methods. A dataset 12,446 Kaggle was used. Pre-processing involved median filtering erosion, followed by segmentation gradient vector flow. Features mean, contrast, entropy were selected Butterfly Optimization Algorithm (BOA) classified with XG Boost, achieving 98% accuracy. Four DL models—Modified CNN, VGG-16, LSTM, Bi-LSTM—were compared. Bi-LSTM outperformed 99% The BOA+XG Boost combination also surpassed standard showing improved accuracy, sensitivity, F1 score. These results suggest models, especially Bi-LSTM, effective for detecting stones.

Language: Английский

---