Synergistic Combination of Machine Learning and Evolutionary and Heuristic Algorithms for Handling Imbalance in Biological and Biomedical Datasets DOI

Sonal Modak,

Mayur Pandya, Patrick Siarry

et al.

Computational intelligence methods and applications, Journal Year: 2024, Volume and Issue: unknown, P. 323 - 362

Published: Jan. 1, 2024

Language: Английский

Reviewing Multimodal Machine Learning and Its Use in Cardiovascular Diseases Detection DOI Open Access
Mohammad Moshawrab, Mehdi Adda,

Abdenour Bouzouane

et al.

Electronics, Journal Year: 2023, Volume and Issue: 12(7), P. 1558 - 1558

Published: March 26, 2023

Machine Learning (ML) and Deep (DL) are derivatives of Artificial Intelligence (AI) that have already demonstrated their effectiveness in a variety domains, including healthcare, where they now routinely integrated into patients’ daily activities. On the other hand, data heterogeneity has long been key obstacle AI, ML DL. Here, Multimodal (Multimodal ML) emerged as method enables training complex DL models use heterogeneous learning process. In addition, integration multiple search for single, comprehensive solution to problem. this review, technical aspects discussed, definition technology its underpinnings, especially fusion. It also outlines differences between others, such Ensemble Learning, well various workflows can be followed ML. article examines depth detection prediction Cardiovascular Diseases, highlighting results obtained so far possible starting points improving aforementioned field. Finally, number most common problems hindering development potential solutions could pursued future studies outlined.

Language: Английский

Citations

24

A Comparative Analysis of Early and Late Fusion for the Multimodal Two-Class Problem DOI Creative Commons

Luis Manuel Pereira,

Addisson Salazar, Luis Vergara

et al.

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 84283 - 84300

Published: Jan. 1, 2023

In this article we carry out a comparison between early (feature) and late (score) multimodal fusion, for the two-class problem. The is made first from general perspective, then specific mathematical analysis. Thus, deduce error probability expressions uncorrelated correlated multivariate Gaussian distribution, assuming perfect model knowledge (Bayes rates). We also corresponding when to be learned finite training set, demonstrating its convergence Bayes rates as set size goes infinite. These demonstrates that fusion best option with knowledge, both degrades due set. This degradation showed grater dimensionality increase of feature space, so, eventually, could better in practical setting. analysis suggests convenience using a, so called, convergence factor , quantify if appropriate close enough rate. Different simulated experiments have been verify validity analysis, well possible extension non-Gaussian models.

Language: Английский

Citations

24

Analyzing mixed construction and demolition waste in material recovery facilities: Evolution, challenges, and applications of computer vision and deep learning DOI Creative Commons

Adrian Langley,

Matthew Lonergan,

Tao Huang

et al.

Resources Conservation and Recycling, Journal Year: 2025, Volume and Issue: 217, P. 108218 - 108218

Published: Feb. 28, 2025

Language: Английский

Citations

1

Cell2Sentence: Teaching Large Language Models the Language of Biology DOI Creative Commons
Daniel Lévine, Syed Asad Rizvi, Sacha Lévy

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: Sept. 14, 2023

Abstract We introduce Cell2Sentence (C2S), a novel method to directly adapt large language models biological context, specifically single-cell transcriptomics. By transforming gene expression data into “cell sentences,” C2S bridges the gap between natural processing and biology. demonstrate cell sentences enable fine-tuning of for diverse tasks in biology, including generation, complex cell-type annotation, direct data-driven text generation. Our experiments reveal that GPT-2, when fine-tuned with C2S, can generate biologically valid cells based on type inputs, accurately predict types from sentences. This illustrates models, through fine-tuning, acquire significant understanding biology while maintaining robust generation capabilities. offers flexible, accessible framework integrate transcriptomics, utilizing existing libraries wide range applications.

Language: Английский

Citations

15

Water Flow Modeling and Forecast in a Water Branch of Mexico City through ARIMA and Transfer Function Models for Anomaly Detection DOI Open Access
David Barrientos, Erick Axel Martinez-Ríos, Sergio A. Navarro-Tuch

et al.

Water, Journal Year: 2023, Volume and Issue: 15(15), P. 2792 - 2792

Published: Aug. 2, 2023

Early identification of anomalies (such as leakages or sensor failures) in urban water distribution systems is critical to mitigating scarcity cities and a challenge resource management. Several data-driven methods based on machine learning algorithms have been proposed the literature for leakage detection systems. Still, most them are challenging implement due their complexity requirements vast amounts reliable data proper model generation. In addition, required infrastructure instrumentation collect needed train models could be unaffordable. This paper presents use comparison Autoregressive Integrated Moving Average Transfer Function generated via Box–Jenkins approach modeling flow anomaly detection. The were fit using from tanks operating branch system Mexico City. results showed that both helped select best type each variable analyzed branch, with Seasonal ARIMA achieving lower mean absolute percentage error than fitted models. Furthermore, this methodology can adjusted different time windows generate alerts at rates does not require large sample size. improve efficiency by detecting such wrong measurements leakages.

Language: Английский

Citations

10

An fNIRS representation and fNIRS-scales multimodal fusion method for auxiliary diagnosis of amnestic mild cognitive impairment DOI Creative Commons
Shiyu Cheng, Pan Shang, Yingwei Zhang

et al.

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 96, P. 106646 - 106646

Published: July 18, 2024

Amnestic mild cognitive impairment (aMCI) is the prodromal period of more serious neurodegenerative diseases (e.g., Alzheimer's disease), characterized by declines in memory and thinking abilities. Auxiliary assessment early diagnosis aMCI are crucial preventing continued deterioration abilities; nevertheless, this task poses a formidable challenge due to inconspicuous nature symptoms. Functional near-infrared spectroscopy (fNIRS) non-invasive, low-cost, user-friendly neuroimaging technique, which capable detecting subtle changes brain activity among different subjects. Moreover, multimodal fusion can assess cognition status from perspectives enhance auxiliary accuracy significantly. This paper proposes an fNIRS representation fNIRS-scales method for aMCI. Specifically, we convert one-dimensional time-series signals into two-dimensional images with Gramian Angular Field achieve end-to-end convolutional neural network. Then, integrate extracted features scales at decision-making level improve aMCI, employing data balance strategy prevent biased prediction. What more, based on features, also propose data-driven scales-screening help physician higher efficiency. We conducted experiments 86 subjects (including 53 patients 33 normal controls) recruited Foshan First People's Hospital. The reaches 88.02% 93.90% further fusion, respectively. With scales-screening, delete 50% scales, reducing test time but only losing 2.54% accuracy.

Language: Английский

Citations

4

Review of Multimodal Data Fusion in Machine Learning DOI Open Access
Leena Arya, Yogesh Kumar Sharma,

Smitha

et al.

Published: Jan. 13, 2025

In many fields, data on the same phenomenon can be derived from a variety of sources, including different detectors, conditions, experiments, and subjects. this article, we refer to every one acquisitions mentioned above frames as "modality." It is quite uncommon for method provide thorough comprehension issue under research, considering diversity natural events. There are concerns beyond employing each modality individually when growing number techniques information an identical framework. generating further levels flexibility. Many these issues "challenges," describe them, cross-disciplinary. The two main topics that article addresses "The reason why combination required" "In what ways accomplish this?" Inspired by instances in areas science technology, primary followed structure based mathematics exhibiting some benefits fusion. We address second concern defining "diversity" applying methods make use tensor well matrix subdivisions. These approaches demonstrate their ability account variability across databases. This aims readers with overview scope, possibilities, promise topic, regardless prior knowledge.

Language: Английский

Citations

0

EHR-based prediction modelling meets multimodal deep learning: A systematic review of structured and textual data fusion methods DOI
Ariel Soares Teles, Ivan Rodrigues de Moura, Francisco Silva

et al.

Information Fusion, Journal Year: 2025, Volume and Issue: unknown, P. 102981 - 102981

Published: Feb. 1, 2025

Language: Английский

Citations

0

A scalable multi-modal learning fruit detection algorithm for dynamic environments DOI Creative Commons
Liang Mao, Zihao Guo, Mingzhe Liu

et al.

Frontiers in Neurorobotics, Journal Year: 2025, Volume and Issue: 18

Published: Feb. 7, 2025

To enhance the detection of litchi fruits in natural scenes, address challenges such as dense occlusion and small target identification, this paper proposes a novel multimodal method, denoted YOLOv5-Litchi. Initially, Neck layer network YOLOv5s is simplified by changing its FPN+PAN structure to an FPN increasing number heads from 3 5. Additionally, with resolutions 80 × pixels 160 are replaced TSCD model's ability detect targets. Subsequently, positioning loss function EIoU function, confidence substituted VFLoss further improve accuracy bounding box reduce missed rate occluded A sliding slice method then employed predict image targets, thereby reducing miss Experimental results demonstrate that proposed model improves accuracy, recall, mean average precision (mAP) 9.5, 0.9, 12.3 percentage points, respectively, compared original model. When benchmarked against other models YOLOx, YOLOv6, YOLOv8, AP value increases 4.0, 6.3, 3.7 respectively. The improved exhibits distinct improvements, primarily focusing on enhancing recall value, which exhibiting reduced targets more accurate prediction frame, indicating suitability for fruit detection. Therefore, significantly enhances mature effectively addresses detection, providing crucial technical support subsequent yield estimation.

Language: Английский

Citations

0

Novel pre‐spatial data fusion deep learning approach for multimodal volumetric outcome prediction models in radiotherapy DOI Creative Commons
John C. Asbach, Anurag Kumar Singh, Austin J. Iovoli

et al.

Medical Physics, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 10, 2025

Given the recent increased emphasis on multimodal neural networks to solve complex modeling tasks, problem of outcome prediction for a course treatment can be framed as fundamentally in nature. A patient's response will vary based their specific anatomy and proposed plan-these factors are spatial closely related. However, additional may also have importance, such non-spatial descriptive clinical characteristics, which structured tabular data. It is critical provide models with comprehensive patient representation possible, but inputs differing data structures incompatible raw form; traditional that consider these require feature engineering prior modeling. In networks, organically integrated into model itself, under one governing optimization, rather than performed prescriptively beforehand. native incompatibility different must addressed. Methods reconcile structural incompatibilities called fusion. We present novel joint early pre-spatial (JEPS) fusion technique demonstrate differences approach produce significant performance even when identical. To volumetric its impact pretreatment overall survival (OS). From retrospective cohort 531 head neck patients treated at our clinic, we prepared an OS dataset 222 data-complete cases 2-year post-treatment time threshold. Each included CT imaging, dose array, approved structure set, summary demographics survey establish single-modality baselines, fit both Cox Proportional Hazards (CPH) dense network only data, then trained 3D convolutional (CNN) volume Then, five competing architectures modalities: two models, late model, JEPS, where merged training upstream most convolution operations. used standardized 10-fold cross validation directly compare all identical train/test splits patients, using area receiver-operator curve (AUC) primary metric. two-tailed Student t-test assess statistical significance (p-value threshold 0.05) any observed differences. The JEPS design scored highest, achieving mean AUC 0.779 ± 0.080. clinical-only CPH second third highest 0.746 0.066 0.720 0.091 AUC, respectively. between three were not statistically significant. All other comparison significantly worse top performing model. For evaluation, architecture achieves better integration improves predictive over common approaches. easily applied CNN.

Language: Английский

Citations

0