Novel large empirical study of deep transfer learning for COVID-19 classification based on CT and X-ray images DOI Creative Commons

Mansour Almutaani,

Turki Turki, Y‐h. Taguchi

et al.

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

Published: Nov. 3, 2024

The early and highly accurate prediction of COVID-19 based on medical images can speed up the diagnostic process thereby mitigate disease spread; therefore, developing AI-based models is an inevitable endeavor. presented work, to our knowledge, first expand model space identify a better performing among 10,000 constructed deep transfer learning (DTL) as follows. First, we downloaded processed 4481 CT X-ray pertaining non-COVID-19 patients, obtained from Kaggle repository. Second, provide inputs four pre-trained (ConvNeXt, EfficientNetV2, DenseNet121, ResNet34) more than million ImageNet database, in which froze convolutional pooling layers feature extraction part while unfreezing training densely connected classifier with Adam optimizer. Third, generate take majority vote two, three, combinations DTL models, resulting $$\sum\nolimits_{r = 2}^{4} {\left( {\begin{array}{*{20}c} 4 \\ r \end{array} } \right)} 11$$ models. Then, combine 11 followed by consecutively generating taking 2}^{11} {11} 2036$$ Finally, select $$7953$$ $$\left( {2036} 2 \right).$$ Experimental results whole datasets using five-fold cross-validation demonstrate that best generated model, named HC, achieving AUC 0.909 when applied dataset, ConvNeXt yielded higher marginal 0.933 compared 0.93 for HX considering dataset. These promising set foundation promoting large generation (LGM) AI.

Language: Английский

Star anise (Illicium verum Hook. F.) polysaccharides: Potential therapeutic management for obesity, hypertension, and diabetes DOI
Abu Hurairah Darwisy Alias, Muhammad Hakimin Shafie

Food Chemistry, Journal Year: 2024, Volume and Issue: 460, P. 140533 - 140533

Published: July 20, 2024

Language: Английский

Citations

5
Novel Automatic Classification of Human Adult Lung Alveolar Type II Cells Infected with SARS-CoV-2 through Deep Transfer Learning Approach DOI Open Access
Turki Turki,

Sarah Al Habib,

Y‐h. Taguchi

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: April 22, 2024

Abstract SARS-CoV-2 can infect alveoli, inducing a lung injury and thereby impairing the function. Healthy alveolar type II (AT2) cells play major role in repair as well keeping alveoli space free from fluids, which is not case for infected AT2 cells. Unlike previous studies, this novel study aims to automatically differentiate between healthy with through using efficient AI-based models, aid disease control treatment. Therefore, we introduce highly accurate deep transfer learning (DTL) approach that works follows. First, downloaded processed 286 images pertaining human (hAT2) cells, obtained electron microscopy public image archive. Second, provided two DTL computations induce ten models. The first computation employs five pre-trained models (including DenseNet201 ResNet152V2) trained on more than million ImageNet database extract features hAT2 images. Then, flattening providing output feature vectors densely connected classifier Adam optimizer. second similar manner minor difference freeze layers extraction while unfreezing training next layers. Compared TFtDenseNet201, experimental results five-fold cross-validation demonstrate TFeDenseNet201 12.37 × faster superior, yielding highest average ACC of 0.993 (F1 0.992 MCC 0.986) statistical significance ( p < 2.2 10 −16 t -test).

Language: Английский

Citations

0
Novel Automatic Classification of Human Adult Lung Alveolar Type II Cells Infected with SARS-CoV-2 through the Deep Transfer Learning Approach DOI Creative Commons
Turki Turki,

Sarah Al Habib,

Y‐h. Taguchi

et al.

Mathematics, Journal Year: 2024, Volume and Issue: 12(10), P. 1573 - 1573

Published: May 17, 2024

Transmission electron microscopy imaging provides a unique opportunity to inspect the detailed structure of infected lung cells with SARS-CoV-2. Unlike previous studies, this novel study aims investigate COVID-19 classification at cellular level in response Particularly, differentiating between healthy and human alveolar type II (hAT2) Hence, we explore feasibility deep transfer learning (DTL) introduce highly accurate approach that works as follows: First, downloaded processed 286 images pertaining hAT2 obtained from public image archive. Second, provided two DTL computations induce ten models. The first computation employs five pre-trained models (including DenseNet201 ResNet152V2) trained on more than one million ImageNet database extract features images. Then, it flattens output feature vectors trained, densely connected classifier Adam optimizer. second similar manner, minor difference freeze layers for extraction while unfreezing jointly training next layers. results using five-fold cross-validation demonstrated TFeDenseNet201 is 12.37× faster superior, yielding highest average ACC 0.993 (F1 0.992 MCC 0.986) statistical significance (P<2.2×10−16 t-test) compared an 0.937 0.938 0.877) counterpart (TFtDenseNet201), showing no (P=0.093 t-test).

Language: Английский

Citations

0
LGM: Novel large empirical study of deep transfer learning for COVID-19 classification based on CT and X-ray images DOI Open Access

Mansour Almutaani,

Turki Turki, Y‐h. Taguchi

et al.

medRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 9, 2024

Abstract The early and highly accurate prediction of COVID-19 based on medical images can speed up the diagnostic process thereby mitigate disease spread; therefore, developing AI-based models is an inevitable endeavor. presented work, to our knowledge, first expand model space identify a better performing among 10000 constructed deep transfer learning (DTL) as follows. First, we downloaded processed 4481 CT X-ray pertaining non-COVID-19 patients, obtained from Kaggle repository. Second, provide inputs four pre-trained (ConvNeXt, EfficientNetV2, DenseNet121, ResNet34) more than million ImageNet database, in which froze convolutional pooling layers feature extraction part while unfreezing training densely connected classifier with Adam optimizer. Third, generate take majority vote two, three, combinations DTL models, resulting models. Then, combine 11 followed by consecutively generating taking Finally, select 7953 . Experimental results whole datasets using five-fold cross-validation demonstrate that best generated model, named HC, achieving AUC 0.909 when applied dataset, ConvNeXt yielded higher marginal 0.933 compared 0.93 for HX considering dataset. These promising set foundation promoting large generation (LGM) AI.

Language: Английский

Citations

0
Novel large empirical study of deep transfer learning for COVID-19 classification based on CT and X-ray images DOI Creative Commons

Mansour Almutaani,

Turki Turki, Y‐h. Taguchi

et al.

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

Published: Nov. 3, 2024

The early and highly accurate prediction of COVID-19 based on medical images can speed up the diagnostic process thereby mitigate disease spread; therefore, developing AI-based models is an inevitable endeavor. presented work, to our knowledge, first expand model space identify a better performing among 10,000 constructed deep transfer learning (DTL) as follows. First, we downloaded processed 4481 CT X-ray pertaining non-COVID-19 patients, obtained from Kaggle repository. Second, provide inputs four pre-trained (ConvNeXt, EfficientNetV2, DenseNet121, ResNet34) more than million ImageNet database, in which froze convolutional pooling layers feature extraction part while unfreezing training densely connected classifier with Adam optimizer. Third, generate take majority vote two, three, combinations DTL models, resulting $$\sum\nolimits_{r = 2}^{4} {\left( {\begin{array}{*{20}c} 4 \\ r \end{array} } \right)} 11$$ models. Then, combine 11 followed by consecutively generating taking 2}^{11} {11} 2036$$ Finally, select $$7953$$ $$\left( {2036} 2 \right).$$ Experimental results whole datasets using five-fold cross-validation demonstrate that best generated model, named HC, achieving AUC 0.909 when applied dataset, ConvNeXt yielded higher marginal 0.933 compared 0.93 for HX considering dataset. These promising set foundation promoting large generation (LGM) AI.

Language: Английский

Citations

0