Segmentation and classification of skin lesions using hybrid deep learning method in the Internet of Medical Things

Skin Research and Technology, Год журнала: 2023, Номер 29(11)

Опубликована: Ноя. 1, 2023

Particularly within the Internet of Medical Things (IoMT) context, skin lesion analysis is critical for precise diagnosis. To improve accuracy and efficiency analysis, CAD systems play a crucial role. segment classify lesions from dermoscopy images, this study focuses on using hybrid deep learning techniques.

Язык: Английский

---

Autism spectrum disorder detection using facial images: A performance comparison of pretrained convolutional neural networks

Healthcare Technology Letters, Год журнала: 2024, Номер 11(4), С. 227 - 239

Опубликована: Янв. 8, 2024

Abstract Autism spectrum disorder (ASD) is a complex psychological syndrome characterized by persistent difficulties in social interaction, restricted behaviours, speech, and nonverbal communication. The impacts of this the severity symptoms vary from person to person. In most cases, ASD appear at age 2 5 continue throughout adolescence into adulthood. While cannot be cured completely, studies have shown that early detection can assist maintaining behavioural development children. Experts are currently studying various machine learning methods, particularly convolutional neural networks, expedite screening process. Convolutional networks considered promising frameworks for diagnosis ASD. This study employs different pre‐trained such as ResNet34, ResNet50, AlexNet, MobileNetV2, VGG16, VGG19 diagnose compared their performance. Transfer was applied every model included achieve higher results than initial models. proposed ResNet50 achieved highest accuracy, 92%, other transfer method also outperformed state‐of‐the‐art models terms accuracy computational cost.

Язык: Английский

---

Evidence for the gut‐skin axis: Common genetic structures in inflammatory bowel disease and psoriasis

Skin Research and Technology, Год журнала: 2024, Номер 30(2)

Опубликована: Фев. 1, 2024

Abstract Background Inflammatory bowel disease (IBD) and psoriasis (Ps) are common immune‐mediated diseases that exhibit clinical comorbidity, possibly due to a genetic structure. However, the exact mechanism remains unknown. Methods The study population consisted of IBD Ps genome‐wide association (GWAS) data. Genetic correlations were first evaluated. Then, overall evaluation employed LD score regression (LDSC), while local assessment utilized heritability estimation from summary statistics (HESS). Causality was conducted through two‐sample Mendelian randomization (2SMR), overlap analysis conditional false discovery rate/conjunctional FDR (cond/conjFDR) method. Finally, LDSC applied specifically expressed genes (LDSC‐SEG) performed at tissue level. For Ps‐specific genes, correlation, causality, shared genetics, trait‐specific associated tissues methodically examined. Results At genomic level, both found between Ps. MR indicated positive causal relationship IBD. conjFDR with threshold < 0.01 identified 43 loci Subsequent investigations into disease‐associated close whole blood, lung, spleen, EBV‐transformed lymphocytes. Conclusion current research offers novel perspective on It contributes an enhanced comprehension structure mechanisms comorbidities in diseases.

Язык: Английский

---

AML‐Net: Attention‐based multi‐scale lightweight model for brain tumour segmentation in internet of medical things

CAAI Transactions on Intelligence Technology, Год журнала: 2024, Номер unknown

Опубликована: Янв. 17, 2024

Abstract Brain tumour segmentation employing MRI images is important for disease diagnosis, monitoring, and treatment planning. Till now, many encoder‐decoder architectures have been developed this purpose, with U‐Net being the most extensively utilised. However, these require a lot of parameters to train semantic gap. Some work tried make lightweight model do channel pruning that made small receptive field which compromised accuracy. The authors propose an attention‐based multi‐scale called AML‐Net in Internet Medical Things overcome above issues. This consists three are trained different scale input along previously learned features diminish loss. Moreover, designed attention module replaced traditional skip connection. For module, six experiments were conducted, from dilated convolution spatial performed well. has convolutions relatively large followed by extract global context encoder low‐level features. Then fine combined decoder's same layer high‐level perform experiment on low‐grade‐glioma dataset provided Cancer Genome Atlas at least Fluid‐Attenuated Inversion Recovery modality. proposed 1/43.4, 1/30.3, 1/28.5, 1/20.2 1/16.7 fewer than Z‐Net, U‐Net, Double BCDU‐Net CU‐Net respectively. authors’ gives results IoU = 0.834, F 1‐score 0.909 sensitivity 0.939, greater CU‐Net, RCA‐IUnet PMED‐Net.

Язык: Английский

---

Classification of Alzheimer disease using DenseNet-201 based on deep transfer learning technique

PLoS ONE, Год журнала: 2024, Номер 19(9), С. e0304995 - e0304995

Опубликована: Сен. 6, 2024

Alzheimer's disease (AD) is a brain illness that causes gradual memory loss. AD has no treatment and cannot be cured, so early detection critical. Various diagnosis approaches are used in this regard, but Magnetic Resonance Imaging (MRI) provides the most helpful neuroimaging tool for detecting AD. In paper, we employ DenseNet-201 based transfer learning technique diagnosing different stages as Non-Demented (ND), Moderate Demented (MOD), Mild (MD), Very (VMD), Severe (SD). The suggested method dataset of MRI scans divided into five classes. Data augmentation methods were to expand size increase DenseNet-201's accuracy. It was found proposed strategy very high classification This practical reliable model delivers success rate 98.24%. findings experiments demonstrate deep approach more accurate performs well compared existing techniques state-of-the-art methods.

Язык: Английский

---

Enhanced Steganalysis for Color Images Using Curvelet Features and Support Vector Machine

Computers, materials & continua/Computers, materials & continua (Print), Год журнала: 2024, Номер 78(1), С. 1311 - 1328

Опубликована: Янв. 1, 2024

Algorithms for steganography are methods of hiding data transfers in media files.Several machine learning architectures have been presented recently to improve stego image identification performance by using spatial information, and these made it feasible handle a wide range problems associated with analysis.Images little information or low payload used embedding methods, but the goal all contemporary research is employ high-payload images classification.To address need both low-and images, this work provides machine-learning approach classification that uses Curvelet transformation efficiently extract characteristics from type images.Support Vector Machine (SVM), commonplace technique, has employed determine whether cover.The Wavelet Obtained Weights (WOW), Spatial Universal Relative Distortion (S-UNIWARD), Highly Undetectable Steganography (HUGO), Minimizing Power Optimal Detector (MiPOD) techniques variety experimental scenarios evaluate proposed method.Using WOW at several payloads, proves its accuracy 98.60%.It exhibits superiority over SOTA methods.

Язык: Английский

---

WaveSeg‐UNet model for overlapped nuclei segmentation from multi‐organ histopathology images

CAAI Transactions on Intelligence Technology, Год журнала: 2024, Номер unknown

Опубликована: Июнь 3, 2024

Abstract Nuclei segmentation is a challenging task in histopathology images. It due to the small size of objects, low contrast, touching boundaries, and complex structure nuclei. Their counting play an important role cancer identification its grading. In this study, WaveSeg‐UNet, lightweight model, introduced segment cancerous nuclei having boundaries. Residual blocks are used for feature extraction. Only one extractor block each level encoder decoder. Normally, images degrade quality lose information during down‐sampling. To overcome loss, discrete wavelet transform (DWT) alongside max‐pooling down‐sampling process. Inverse DWT regenerate original up‐sampling. bottleneck proposed atrous spatial channel pyramid pooling (ASCPP) extract effective high‐level features. The ASCPP modified layers increase area receptive field. Spatial channel‐based attention focus on location class identified objects. Finally, watershed as post processing technique identify refine boundaries counted facilitate pathologists. same domain transfer learning retrain model adaptability. Results compared with state‐of‐the‐art models, it outperformed existing studies.

Язык: Английский

---

Improving Skin Lesion Diagnosis: Hybrid Blur Detection for Accurate Dermatological Image Analysis

Communications in computer and information science, Год журнала: 2024, Номер unknown, С. 225 - 240

Опубликована: Янв. 1, 2024

Язык: Английский

---

Intelligent skin lesion segmentation using deformable attention Transformer U‐Net with bidirectional attention mechanism in skin cancer images

Skin Research and Technology, Год журнала: 2024, Номер 30(8)

Опубликована: Авг. 1, 2024

Abstract Background In recent years, the increasing prevalence of skin cancers, particularly malignant melanoma, has become a major concern for public health. The development accurate automated segmentation techniques lesions holds immense potential in alleviating burden on medical professionals. It is substantial clinical importance early identification and intervention cancer. Nevertheless, irregular shape, uneven color, noise interference have presented significant challenges to precise segmentation. Therefore, it crucial develop high‐precision intelligent lesion framework treatment. Methods A precision‐driven model cancer images proposed based Transformer U‐Net, called BiADATU‐Net, which integrates deformable attention bidirectional blocks into U‐Net. encoder part utilizes with dual block, allowing adaptive learning global local features. decoder incorporates specifically tailored scSE modules within skip connection layers capture image‐specific context information strong feature fusion. Additionally, convolution aggregated two different learn features prediction. Results series experiments are conducted four image datasets (i.e., ISIC2016, ISIC2017, ISIC2018, PH2). findings show that our exhibits satisfactory performance, all achieving an accuracy rate over 96%. Conclusion Our experiment results validate BiADATU‐Net achieves competitive performance supremacy compared some state‐of‐the‐art methods. valuable field

Язык: Английский

---

Improving security performance of Internet of Medical Things using hybrid metaheuristic model

Multimedia Tools and Applications, Год журнала: 2024, Номер unknown

Опубликована: Апрель 18, 2024

Язык: Английский

---