A novel multi class disease detection of chest x-ray images using deep learning with pre trained transfer learning models for medical imaging applications

Research Square (Research Square), Год журнала: 2024, Номер unknown

Опубликована: Фев. 14, 2024

Abstract Images from chest X-rays (CXR) are thought to help observe and research various kinds of pulmonary illnesses. Several works were suggested in the literature for recognizing unique lung diseases, only a few studies focused on developing model identify joint classes diseases. A patient with negative diagnosis one condition may have other disease, vice versa. However, since many illnesses lung-related, can multiple simultaneously. This paper proposes deep learning (DL)-based pre-trained transfer (TL) effectively detecting classifying multiclass diseases CXR images. The system involves five phases: preprocessing, dataset balancing, feature learning, selection, classification. Firstly, images preprocessed by performing filtering, contrast enhancement, data augmentation. After that, balancing is performed using Synthetic Minority Oversampling Technique (SMOTE). Next, features learned spatial channel-attention-based Xception Network (SCAXN). optimal selected nonlinear decreasing inertia weight-based rock hyraxes swarm optimization (NIWRHSO). Finally, classification uses soft sign-incorporated bidirectional gated recurrent unit (SBIGRU). Two public datasets, COVID-19 Radiography (C19RY) Tuberculosis (TB-CXR), been obtained Kaggle, outcomes confirmed that proposed attains superior results prevailing methods.

Язык: Английский

---

Development of a chest X-ray machine learning convolutional neural network model on a budget and using artificial intelligence explainability techniques to analyze patterns of machine learning inference

JAMIA Open, Год журнала: 2024, Номер 7(2)

Опубликована: Апрель 8, 2024

Abstract Objective Machine learning (ML) will have a large impact on medicine and accessibility is important. This study’s model was used to explore various concepts including how varying features of impacted behavior. Materials Methods study built an ML that classified chest X-rays as normal or abnormal by using ResNet50 base with transfer learning. A contrast enhancement mechanism implemented improve performance. After training dataset publicly available radiographs, performance metrics were determined test set. The substituted deeper architectures (ResNet101/152) visualization methods help determine patterns inference. Results Performance accuracy 79%, recall 69%, precision 96%, area under the curve 0.9023. Accuracy improved 82% 74% enhancement. When applied ratio pixels for inference measured, resulted in larger portions image compared ResNet50. Discussion performed par many existing models despite consumer-grade hardware smaller datasets. Individual vary thus single model’s explainability may not be generalizable. Therefore, this varied architecture studied With ResNet architectures, machine make decisions. Conclusion An example custom showed AI (Artificial Intelligence) can accessible hardware, it also demonstrated studying themes architectures.

Язык: Английский

---

Enhancing multi-class lung disease classification in chest x-ray images: A hybrid manta-ray foraging volcano eruption algorithm boosted multilayer perceptron neural network approach

Network Computation in Neural Systems, Год журнала: 2024, Номер unknown, С. 1 - 32

Опубликована: Май 16, 2024

One of the most used diagnostic imaging techniques for identifying a variety lung and bone-related conditions is chest X-ray. Recent developments in deep learning have demonstrated several successful cases illness diagnosis from X-rays. However, issues stability class imbalance still need to be resolved. Hence this manuscript, multi-class disease classification x-ray images using hybrid manta-ray foraging volcano eruption algorithm boosted multilayer perceptron neural network approach proposed (MPNN-Hyb-MRF-VEA). Initially, input X-ray are taken Covid-Chest dataset. Anisotropic diffusion Kuwahara filtering (ADKF) enhance quality these lower noise. To capture significant discriminative features, Term frequency-inverse document frequency (TF-IDF) based feature extraction method utilized case. The Multilayer Perceptron Neural Network (MPNN) serves as model disorders COVID-19, pneumonia, tuberculosis (TB), normal. A Hybrid Manta-Ray Foraging Volcano Eruption Algorithm (Hyb-MRF-VEA) introduced further optimize fine-tune MPNN's parameters. Python platform accurately evaluate methodology. performance provides 23.21%, 12.09%, 5.66% higher accuracy compared with existing methods like NFM, SVM, CNN respectively.

Язык: Английский

---

Adversarially Enhanced Learning (AEL): Robust lightweight deep learning approach for radiology image classification against adversarial attacks

Image and Vision Computing, Год журнала: 2024, Номер unknown, С. 105405 - 105405

Опубликована: Дек. 1, 2024

Язык: Английский

---

COVID-Net Biochem: an explainability-driven framework to building machine learning models for predicting survival and kidney injury of COVID-19 patients from clinical and biochemistry data

Scientific Reports, Год журнала: 2023, Номер 13(1)

Опубликована: Окт. 9, 2023

Abstract Since the World Health Organization declared COVID-19 a pandemic in 2020, global community has faced ongoing challenges controlling and mitigating transmission of SARS-CoV-2 virus, as well its evolving subvariants recombinants. A significant challenge during not only been accurate detection positive cases but also efficient prediction risks associated with complications patient survival probabilities. These tasks entail considerable clinical resource allocation attention. In this study, we introduce COVID-Net Biochem, versatile explainable framework for constructing machine learning models. We apply to predict likelihood developing Acute Kidney Injury hospitalization, utilizing biochemical data transparent, systematic approach. The proposed approach advances model design by seamlessly integrating domain expertise explainability tools, enabling decisions be based on key biomarkers. This fosters more transparent interpretable decision-making process made machines specifically medical applications. More specifically, comprises two phases: first phase, referred “clinician-guided design” dataset is preprocessed using AI expert input. To better demonstrate prepared benchmark carefully curated markers clinician assessments kidney injury patients. was selected from cohort 1366 individuals at Stony Brook University. Moreover, designed trained diverse collection models, encompassing gradient-based boosting tree architectures deep transformer architectures, markers. second called “explainability-driven refinement” employs methods gain deeper understanding each model’s identify overall impact individual bias identification. context, used models constructed previous phase task analyzed outcomes alongside over 8 years experience validity made. explainability-driven insights obtained, conjunction feedback, are then utilized guide refine training policies architectural iteratively. aims enhance performance trustworthiness final Employing framework, attained 93.55% accuracy 88.05% predicting complications. have available through an open-source platform. Although production-ready solution, study serve catalyst scientists, researchers, citizen scientists develop innovative trustworthy decision support solutions, ultimately assisting clinicians worldwide managing outcomes.

Язык: Английский

---