Enhancing Melanoma Diagnosis with Advanced Deep Learning Models Focusing on Vision Transformer, Swin Transformer, and ConvNeXt

Dermatopathology, Год журнала: 2024, Номер 11(3), С. 239 - 252

Опубликована: Авг. 15, 2024

Skin tumors, especially melanoma, which is highly aggressive and progresses quickly to other sites, are an issue in various parts of the world. Nevertheless, one only way save lives detect it at its initial stages. This study explores application advanced deep learning models for classifying benign malignant melanoma using dermoscopic images. The aim enhance accuracy efficiency diagnosis with ConvNeXt, Vision Transformer (ViT) Base-16, Swin V2 Small (Swin S) models. ConvNeXt model, integrates principles both convolutional neural networks transformers, demonstrated superior performance, balanced precision recall metrics. dataset, sourced from Kaggle, comprises 13,900 uniformly sized images, preprocessed standardize inputs Experimental results revealed that achieved highest diagnostic among tested 91.5%, rates 90.45% 92.8% cases, 92.61% 90.2% respectively. F1-scores were 91.61% cases 91.39% cases. research points out potential hybrid architectures medical image analysis, particularly early detection.

Язык: Английский

---

Segmentation for mammography classification utilizing deep convolutional neural network

BMC Medical Imaging, Год журнала: 2024, Номер 24(1)

Опубликована: Дек. 18, 2024

Mammography for the diagnosis of early breast cancer (BC) relies heavily on identification masses. However, in stages, it might be challenging to ascertain whether a mass is benign or malignant. Consequently, many deep learning (DL)-based computer-aided (CAD) approaches BC classification have been developed. Recently, transformer model has emerged as method overcoming constraints convolutional neural networks (CNN). Thus, our primary goal was determine how well an improved could distinguish between and malignant tissues. In this instance, we drew Mendeley data repository's INbreast dataset, which includes types. Additionally, segmentation anything (SAM) used generate optimized cutoff region interest (ROI) extraction from all mammograms. We implemented successful architecture modification at bottom layer pyramid (PTr) identify mammography images. The proposed PTr using transfer (TL) approach with technique achieved best accuracy 99.96% binary classifications area under curve (AUC) score 99.98%, respectively. also compared performance other vision transformers (ViT) DL models, MobileNetV3 EfficientNetB7, study, modified prediction image approaches. Data techniques accurately regions affected by BC. Finally, classified tissues, vital radiologists guide future treatment.

Язык: Английский

---

A novel Skin lesion prediction and classification technique: ViT‐GradCAM

Skin Research and Technology, Год журнала: 2024, Номер 30(9)

Опубликована: Сен. 1, 2024

Skin cancer is one of the highly occurring diseases in human life. Early detection and treatment are prime necessary points to reduce malignancy infections. Deep learning techniques supplementary tools assist clinical experts detecting localizing skin lesions. Vision transformers (ViT) based on image segmentation classification using multiple classes provide fairly accurate gaining more popularity due legitimate multiclass prediction capabilities.

Язык: Английский

---

Ensemble of hybrid model based technique for early detecting of depression based on SVM and neural networks

Scientific Reports, Год журнала: 2024, Номер 14(1)

Опубликована: Окт. 26, 2024

The prevalence of depression has increased dramatically over the last several decades: it is frequently overlooked and can have a significant impact on both physical mental health. Therefore, crucial to develop an automated detection system that instantly identify whether person depressed. Currently, machine learning (ML) artificial neural networks (ANNs) are among most promising approaches for developing computer-based systems predict health issues, such as depression. This study propose ensemble hybrid model-based techniques aims build strong model considers many psychological sociodemographic characteristics individual detect Support vector machines (SVM) multilayer perceptrons (MLP) two fundamental methods used construct suggested approach. DeprMVM served meta-learner. In this study, level-1 learner, whereas SVM MLP level-0 learners. After classifiers trained tested at level 0, their outputs based independent dependent variables in new data set was train meta-classifier. training class imbalance reduced by applying synthetic minority oversampling technique (SMOTE) cluster sampling together, which improved accuracy detecting Additionally, effectively reduce risk over-fitting from simply duplicating points. To further confirm effectiveness proposed method, various performance evaluation metrics were calculated compared with previous studies conducted specific dataset. conclusion, all identifying depression, approach had best accuracy, 99.39%, F1-score 99.51%.

Язык: Английский

---

Automated explainable deep learning framework for multiclass skin cancer detection and classification using hybrid YOLOv8 and vision transformer (ViT)

Biomedical Signal Processing and Control, Год журнала: 2025, Номер 108, С. 107934 - 107934

Опубликована: Апрель 29, 2025

Язык: Английский

---

Enhancing Dermatological Diagnosis Through Medical Image Analysis: How Effective Is YOLO11 Compared to Leading CNN Models?

NDT, Год журнала: 2025, Номер 3(2), С. 11 - 11

Опубликована: Май 21, 2025

Skin diseases represent a major worldwide health hazard affecting millions of people yearly and substantially compromising healthcare systems. Particularly in areas where dermatologists are scarce, standard diagnostic techniques, which mostly rely on visual inspection clinical experience, frequently subjective, time-consuming, prone to mistakes. This investigation undertakes comparative analysis four state-of-the-art deep learning architectures, YOLO11, YOLOv8, VGG16, ResNet50, the context skin disease identification. study evaluates performance these models using pivotal metrics, building upon foundation YOLO paradigm, revolutionized spatial attention multi-scale representation. A properly selected collection 900 high-quality dermatological images with nine categories was used for investigation. Robustness generalizability were guaranteed by data augmentation hyperparameter adjustment. By varying benchmark balancing accuracy recall while limiting false positives negatives, YOLO11 obtained test 80.72%, precision 88.7%, 86.7%, an F1 score 87.0%. The expedition signifies promising trajectory development highly accurate detection models. Our not only highlights strengths weaknesses model but also underscores rapid techniques medical imaging.

Язык: Английский

---

Explainable AI-Based Skin Cancer Detection Using CNN, Particle Swarm Optimization and Machine Learning

Journal of Imaging, Год журнала: 2024, Номер 10(12), С. 332 - 332

Опубликована: Дек. 22, 2024

Skin cancer is among the most prevalent cancers globally, emphasizing need for early detection and accurate diagnosis to improve outcomes. Traditional diagnostic methods, based on visual examination, are subjective, time-intensive, require specialized expertise. Current artificial intelligence (AI) approaches skin face challenges such as computational inefficiency, lack of interpretability, reliance standalone CNN architectures. To address these limitations, this study proposes a comprehensive pipeline combining transfer learning, feature selection, machine-learning algorithms accuracy. Multiple pretrained models were evaluated, with Xception emerging optimal choice its balance efficiency performance. An ablation further validated effectiveness freezing task-specific layers within architecture. Feature dimensionality was optimized using Particle Swarm Optimization, reducing dimensions from 1024 508, significantly enhancing efficiency. Machine-learning classifiers, including Subspace KNN Medium Gaussian SVM, improved classification Evaluated ISIC 2018 HAM10000 datasets, proposed achieved impressive accuracies 98.5% 86.1%, respectively. Moreover, Explainable-AI (XAI) techniques, Grad-CAM, LIME, Occlusion Sensitivity, enhanced interpretability. This approach provides robust, efficient, interpretable solution automated in clinical applications.

Язык: Английский

---

Analysis of the use of digital technologies in the preliminary diagnosis of dermatological diseases: a systematic review

Archives of Dermatological Research, Год журнала: 2024, Номер 317(1)

Опубликована: Дек. 20, 2024

Язык: Английский

---