Prognosis prediction in traumatic brain injury patients using machine learning algorithms

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

Опубликована: Янв. 18, 2023

Predicting treatment outcomes in traumatic brain injury (TBI) patients is challenging worldwide. The present study aimed to achieve the most accurate machine learning (ML) algorithms predict of TBI by evaluating demographic features, laboratory data, imaging indices, and clinical features. We used data from 3347 admitted a tertiary trauma centre Iran 2016 2021. After exclusion incomplete 1653 remained. ML such as random forest (RF) decision tree (DT) with ten-fold cross-validation develop best prediction model. Our findings reveal that among different variables included this study, motor component Glasgow coma scale, condition pupils, cisterns were reliable features for predicting in-hospital mortality, while patients' age takes place when considering long-term survival patients. Also, we found RF algorithm model short-term mortality However, generalized linear (GLM) showed performance (with an accuracy rate 82.03 ± 2.34) results using appropriate markers further development, has potential short- long-term.

Язык: Английский

---

Parameter adaptive unit-linking pulse coupled neural network based MRI–PET/SPECT image fusion

Biomedical Signal Processing and Control, Год журнала: 2023, Номер 83, С. 104659 - 104659

Опубликована: Фев. 3, 2023

Язык: Английский

---

Brain tumor detection and screening using artificial intelligence techniques: Current trends and future perspectives

Computers in Biology and Medicine, Год журнала: 2023, Номер 163, С. 107063 - 107063

Опубликована: Июнь 1, 2023

Язык: Английский

---

Multimodal diagnosis model of Alzheimer’s disease based on improved Transformer

BioMedical Engineering OnLine, Год журнала: 2024, Номер 23(1)

Опубликована: Янв. 19, 2024

Abstract Purpose Recent technological advancements in data acquisition tools allowed neuroscientists to acquire different modality diagnosis Alzheimer’s disease (AD). However, how fuse these enormous amount improve recognizing rate and find significance brain regions is still challenging. Methods The algorithm used multimodal medical images [structural magnetic resonance imaging (sMRI) positron emission tomography (PET)] as experimental data. Deep feature representations of sMRI PET are extracted by 3D convolution neural network (3DCNN). An improved Transformer then progressively learn global correlation information among features. Finally, the from modalities fused for identification. A model-based visualization method explain decisions model identify related AD. Results attained a noteworthy classification accuracy 98.1% (AD) using Disease Neuroimaging Initiative (ADNI) dataset. Upon examining results, distinct associated with AD were observed across image modalities. Notably, left parahippocampal region emerged consistently prominent significant area. Conclusions large number comparative experiments have been carried out model, results verify reliability model. In addition, adopts analysis based on characteristics which improves interpretability Some disease-related found provides reliable clinical research.

Язык: Английский

---

Fuzzy Deep Learning for the Diagnosis of Alzheimer's Disease: Approaches and Challenges

IEEE Transactions on Fuzzy Systems, Год журнала: 2024, Номер 32(10), С. 5477 - 5492

Опубликована: Июнь 18, 2024

Язык: Английский

---

Alzheimer’s disease diagnosis from single and multimodal data using machine and deep learning models: Achievements and future directions

Expert Systems with Applications, Год журнала: 2024, Номер 255, С. 124780 - 124780

Опубликована: Июль 14, 2024

Язык: Английский

---

A comprehensive survey of deep learning-based lightweight object detection models for edge devices

Artificial Intelligence Review, Год журнала: 2024, Номер 57(9)

Опубликована: Авг. 10, 2024

This study concentrates on deep learning-based lightweight object detection models edge devices. Designing such recognition is more difficult than ever due to the growing demand for accurate, quick, and low-latency various The most recent methods are comprehensively described in this work. Information backbone architectures used by these detectors has been listed. training inference processes concerning learning applications devices being discussed. To raise readers' awareness of developing domain, a variety related utilities have offered. potent, based suggested as counter problems. On well-known datasets MS-COCO PASCAL-VOC, we thoroughly examine performance certain conventional detectors.

Язык: Английский

---

GATransformer: A Graph Attention Network-Based Transformer Model to Generate Explainable Attentions for Brain Tumor Detection

Algorithms, Год журнала: 2025, Номер 18(2), С. 89 - 89

Опубликована: Фев. 6, 2025

Brain tumors profoundly affect human health owing to their intricacy and the difficulties associated with early identification treatment. Precise diagnosis is essential for effective intervention; nevertheless, resemblance among tumor forms often complicates of brain types, particularly in stages. The latest deep learning systems offer very high classification accuracy but lack explainability help patients understand prediction process. GATransformer, a graph attention network (GAT)-based Transformer, uses mechanism, GAT, Transformer identify preserve key neural channels. channel module extracts deeper properties from weight-channel connections improve model representation. Integrating these elements results reduction size enhancement computing efficiency, while preserving adequate performance. proposed assessed using two publicly accessible datasets, FigShare Kaggle, cross-validated BraTS2019 BraTS2020 demonstrating explainability. Notably, GATransformer generates interpretable maps, visually highlighting regions aid clinical understanding medical imaging.

Язык: Английский

---

Phenotyping the Histopathological Subtypes of Non-Small-Cell Lung Carcinoma: How Beneficial Is Radiomics?

Diagnostics, Год журнала: 2023, Номер 13(6), С. 1167 - 1167

Опубликована: Март 18, 2023

The aim of this study was to investigate the usefulness radiomics in absence well-defined standard guidelines. Specifically, we extracted features from multicenter computed tomography (CT) images differentiate between four histopathological subtypes non-small-cell lung carcinoma (NSCLC). In addition, results that varied with model were compared. We investigated presence batch effects and impact feature harmonization on models’ performance. Moreover, question how training dataset composition influenced selected subsets and, consequently, model’s performance also investigated. Therefore, through combining data two publicly available datasets, involves a total 152 squamous cell (SCC), 106 large (LCC), 150 adenocarcinoma (ADC), 58 no other specified (NOS). Through matRadiomics tool, which is an example Image Biomarker Standardization Initiative (IBSI) compliant software, 1781 each malignant lesions identified CT images. After analysis harmonization, based ComBat tool integrated matRadiomics, datasets (the harmonized non-harmonized) given as input machine learning modeling pipeline. following steps articulated: (i) training-set/test-set splitting (80/20); (ii) Kruskal–Wallis LASSO linear regression for selection; (iii) training; (iv) validation hyperparameter optimization; (v) testing. Model optimization consisted 5-fold cross-validated Bayesian optimization, repeated ten times (inner loop). whole pipeline 10 (outer loop) six different classification algorithms. stability selection evaluated. Results showed present even if voxels resampled isotropic form whether correctly removed them, though performances decreased. low accuracy (61.41%) reached when differentiating subtypes, high average area under curve (AUC) (0.831). Further, NOS subtype classified almost completely correct (true positive rate ~90%). increased (77.25%) only SCC ADC considered, well AUC (0.821) obtained—although decreased 58%. contributed most those wavelet decomposed Laplacian Gaussian (LoG) filtered they belonged texture class.. conclusion, our affected by effects, could significantly alter performance, them. Although seemed be informative features, absolute subset not since it changed depending training/testing splitting. chosen methods, reach binary tasks, but underperform multiclass problems. It is, therefore, essential scientific community propose more systematic approach, focusing studies, clear solid guidelines facilitate translation clinical practice.

Язык: Английский

---

Empowering precision medicine: AI-driven schizophrenia diagnosis via EEG signals: A comprehensive review from 2002–2023

Applied Intelligence, Год журнала: 2023, Номер 54(1), С. 35 - 79

Опубликована: Дек. 5, 2023

Язык: Английский

---