Skin cancer diagnosis based on deep transfer learning and sparrow search algorithm

Neural Computing and Applications, Journal Year: 2022, Volume and Issue: 35(1), P. 815 - 853

Published: Sept. 23, 2022

Abstract Skin cancer affects the lives of millions people every year, as it is considered most popular form cancer. In USA alone, approximately three and a half million are diagnosed with skin annually. The survival rate diminishes steeply progresses. Despite this, an expensive difficult procedure to discover this type in early stages. study, threshold-based automatic approach for detection, classification, segmentation utilizing meta-heuristic optimizer named sparrow search algorithm (SpaSA) proposed. Five U-Net models (i.e., U-Net, U-Net++, Attention V-net, Swin U-Net) different configurations utilized perform process. Besides SpaSA used optimization hyperparameters using eight pre-trained CNN VGG16, VGG19, MobileNet, MobileNetV2, MobileNetV3Large, MobileNetV3Small, NASNetMobile, NASNetLarge). dataset gathered from five public sources which two types datasets generated 2-classes 10-classes). For segmentation, concerning “skin classification” dataset, best reported scores by U-Net++ DenseNet201 backbone architecture 0.104, $$94.16\%$$ 94.16 % , $$91.39\%$$ 91.39 $$99.03\%$$ 99.03 $$96.08\%$$ 96.08 $$96.41\%$$ 96.41 $$77.19\%$$ 77.19 $$75.47\%$$ 75.47 terms loss, accuracy, F1-score, AUC, IoU, dice, hinge, squared respectively, while “PH2” 0.137, $$94.75\%$$ 94.75 $$92.65\%$$ 92.65 $$92.56\%$$ 92.56 $$92.74\%$$ 92.74 $$96.20\%$$ 96.20 $$86.30\%$$ 86.30 $$69.28\%$$ 69.28 $$68.04\%$$ 68.04 precision, sensitivity, specificity, respectively. “ISIC 2019 2020 Melanoma” overall accuracy applied experiments $$98.27\%$$ 98.27 MobileNet model. Similarly, “Melanoma Classification (HAM10K)” $$98.83\%$$ 98.83 diseases image” $$85.87\%$$ 85.87 MobileNetV2 After computing results, suggested compared 13 related studies.

Language: Английский

---

Enhancement of Detection of Diabetic Retinopathy Using Harris Hawks Optimization with Deep Learning Model

Computational Intelligence and Neuroscience, Journal Year: 2022, Volume and Issue: 2022, P. 1 - 13

Published: May 26, 2022

In today’s world, diabetic retinopathy is a very severe health issue, which affecting many humans of different age groups. Due to the high levels blood sugar, minuscule vessels in retina may get damaged no time and further lead retinal detachment even sometimes glaucoma blindness. If can be diagnosed at early stages, then affected people will not losing their vision also human lives saved. Several machine learning deep methods have been applied on available data sets retinopathy, but they were unable provide better results terms accuracy preprocessing optimizing classification feature extraction process. To overcome issues like optimization existing systems, we considered Diabetic Retinopathy Debrecen Data Set from UCI repository designed model with principal component analysis (PCA) for dimensionality reduction, extract most important features, Harris hawks algorithm used optimize The shown by respect specificity, precision, accuracy, recall are much satisfactory compared systems.

Language: Английский

---

Hierarchical Harris hawks optimizer for feature selection

Journal of Advanced Research, Journal Year: 2023, Volume and Issue: 53, P. 261 - 278

Published: Jan. 20, 2023

Feature selection is a typical NP-hard problem. The main methods include filter, wrapper-based, and embedded methods. Because of its characteristics, the wrapper method must swarm intelligence algorithm, performance in feature closely related to algorithm's quality. Therefore, it essential choose design suitable algorithm improve based on wrapper. Harris hawks optimization (HHO) superb approach that has just been introduced. It high convergence rate powerful global search capability but an unsatisfactory effect dimensional problems or complex problems. we introduced hierarchy HHO's ability deal with selection. To make obtain good accuracy fewer features run faster selection, improved HHO named EHHO. On 30 UCI datasets, (EHHO) can achieve very classification less running time features. We first conducted extensive experiments 23 classical benchmark functions compared EHHO many state-of-the-art metaheuristic algorithms. Then transform into binary (bEHHO) through conversion function verify extraction data sets. Experiments show better speed minimum than other peers. At same time, HHO, significantly weakness dealing functions. Moreover, datasets repository, bEHHO comparative Compared original bHHO, excellent also bHHO time.

Language: Английский

---

Recent Advances in Harris Hawks Optimization: A Comparative Study and Applications

Electronics, Journal Year: 2022, Volume and Issue: 11(12), P. 1919 - 1919

Published: June 20, 2022

The Harris hawk optimizer is a recent population-based metaheuristics algorithm that simulates the hunting behavior of hawks. This swarm-based performs optimization procedure using novel way exploration and exploitation multiphases search. In this review research, we focused on applications developments well-established robust (HHO) as one most popular techniques 2020. Moreover, several experiments were carried out to prove powerfulness effectivness HHO compared with nine other state-of-art algorithms Congress Evolutionary Computation (CEC2005) CEC2017. literature paper includes deep insight about possible future directions ideas worth investigations regarding new variants its widespread applications.

Language: Английский

---

Ensemble deep learning for brain tumor detection

Frontiers in Computational Neuroscience, Journal Year: 2022, Volume and Issue: 16

Published: Sept. 2, 2022

With the quick evolution of medical technology, era big data in medicine is quickly approaching. The analysis and mining these significantly influence prediction, monitoring, diagnosis, treatment tumor disorders. Since it has a wide range traits, low survival rate, an aggressive nature, brain regarded as deadliest most devastating disease. Misdiagnosed tumors lead to inadequate treatment, reducing patient's life chances. Brain detection highly challenging due capacity distinguish between aberrant normal tissues. Effective therapy long-term are made possible for patient by correct diagnosis. Despite extensive research, there still certain limitations detecting because unusual distribution pattern lesions. Finding region with small number lesions can be difficult areas tend look healthy. It directly reduces classification accuracy, extracting choosing informative features challenging. A significant role played automatically classifying early-stage utilizing deep machine learning approaches. This paper proposes hybrid model Convolutional Neural Network-Long Short Term Memory (CNN-LSTM) predicting through Magnetic Resonance Images (MRI). We experiment on MRI image dataset. First, preprocessed efficiently, then, Network (CNN) applied extract from images. proposed predicts accuracy 99.1%, precision 98.8%, recall 98.9%, F1-measure 99.0%.

Language: Английский

---

CovH2SD: A COVID-19 detection approach based on Harris Hawks Optimization and stacked deep learning

Expert Systems with Applications, Journal Year: 2021, Volume and Issue: 186, P. 115805 - 115805

Published: Sept. 5, 2021

Language: Английский

---

Hybrid deep learning and genetic algorithms approach (HMB-DLGAHA) for the early ultrasound diagnoses of breast cancer

Neural Computing and Applications, Journal Year: 2022, Volume and Issue: 34(11), P. 8671 - 8695

Published: Jan. 31, 2022

Language: Английский

---

A vision-based deep learning approach for independent-users Arabic sign language interpretation

Multimedia Tools and Applications, Journal Year: 2022, Volume and Issue: 82(5), P. 6807 - 6826

Published: Aug. 10, 2022

Abstract More than 5% of the people around world are deaf and have severe difficulties in communicating with normal according to World Health Organization (WHO). They face a real challenge express anything without an interpreter for their signs. Nowadays, there lot studies related Sign Language Recognition (SLR) that aims reduce this gap between as it can replace need interpreter. However, challenges facing sign recognition systems such low accuracy, complicated gestures, high-level noise, ability operate under variant circumstances generalize or be locked limitations. Hence, many researchers proposed different solutions overcome these problems. Each language has its signs very challenging cover all languages’ The current study objectives: (i) presenting dataset 20 Arabic words, (ii) proposing deep learning (DL) architecture by combining convolutional neural network (CNN) recurrent (RNN). suggested reported 98% accuracy on presented dataset. It also 93.4% 98.8% top-1 top-5 accuracies UCF-101

Language: Английский

---

A generic optimization and learning framework for Parkinson disease via speech and handwritten records

Journal of Ambient Intelligence and Humanized Computing, Journal Year: 2022, Volume and Issue: 14(8), P. 10673 - 10693

Published: Aug. 26, 2022

Abstract Parkinson’s disease (PD) is a neurodegenerative disorder with slow progression whose symptoms can be identified at late stages. Early diagnosis and treatment of PD help to relieve the delay progression. However, this very challenging due similarities between other diseases. The current study proposes generic framework for using handwritten images (or) speech signals. For handwriting images, 8 pre-trained convolutional neural networks (CNN) via transfer learning tuned by Aquila Optimizer were trained on NewHandPD dataset diagnose PD. signals, features from MDVR-KCL are extracted numerically 16 feature extraction algorithms fed 4 different machine Grid Search algorithm, graphically 5 techniques pretrained CNN structures. authors propose new technique in extracting voice based segmentation variable speech-signal-segment-durations, i.e., use durations phase. Using proposed technique, datasets 281 numerical generated. Results experiments collected recorded. dataset, best-reported metric 99.75% VGG19 structure. metrics 99.94% KNN SVM ML combined features; 100% mel-specgram graphical These results better than state-of-the-art researches.

Language: Английский

---

A comprehensive framework towards segmenting and classifying breast cancer patients using deep learning and Aquila optimizer

Journal of Ambient Intelligence and Humanized Computing, Journal Year: 2023, Volume and Issue: 14(6), P. 7897 - 7917

Published: April 7, 2023

Abstract Breast cancer is among the major frequent types of worldwide, causing a significant death rate every year. It second most prevalent malignancy in Egypt. With increasing number new cases, it vital to diagnose breast its early phases avoid serious complications and deaths. Therefore, routine screening important. current evolution deep learning, medical imaging became one interesting fields. The purpose work suggest hybrid framework for both classification segmentation scans. consists two phases, namely phase phase. In phase, five different CNN architectures via transfer MobileNet, MobileNetV2, NasNetMobile, VGG16, VGG19, are applied. Aquila optimizer used calculation optimal hyperparameters TL architectures. Four datasets representing four modalities (i.e., MRI, Mammographic, Ultrasound images, Histopathology slides) training purposes. can perform binary- multi-class classification. structures, U-Net, Swin Attention U-Net++, V-Net, applied identify region interest ultrasound images. reported results prove efficiency suggested against state-of-the-art studies.

Language: Английский

---