A deep neural network using modified EfficientNet for skin cancer detection in dermoscopic images

Decision Analytics Journal, Journal Year: 2023, Volume and Issue: 8, P. 100278 - 100278

Published: June 25, 2023

Artificial intelligence (AI) systems can assist in analyzing medical images and aiding the early detection of diseases. AI also ensure quality services by avoiding misdiagnosis caused human errors. This study proposes a deep neural network (DNN) model with fine-tuned training improved learning performance on dermoscopic for skin cancer detection. A knowledge base DL models is constructed combining different datasets. Transfer fine-tuning are implemented faster proposed limited dataset. The data augmentation techniques applied to enhance model. total 58,032 refined were used this study. output layered architecture aggregated perform binary classification cancer. trained investigated multiclass tasks. metrics confirm that DNN modified EfficientNetV2-M outperforms state-of-the-art learning-based models.

Language: Английский

---

MSRNet: Multiclass Skin Lesion Recognition Using Additional Residual Block Based Fine-Tuned Deep Models Information Fusion and Best Feature Selection

Diagnostics, Journal Year: 2023, Volume and Issue: 13(19), P. 3063 - 3063

Published: Sept. 26, 2023

Cancer is one of the leading significant causes illness and chronic disease worldwide. Skin cancer, particularly melanoma, becoming a severe health problem due to its rising prevalence. The considerable death rate linked with melanoma requires early detection receive immediate successful treatment. Lesion classification are more challenging many forms artifacts such as hairs, noise, irregularity lesion shape, color, irrelevant features, textures. In this work, we proposed deep-learning architecture for classifying multiclass skin cancer detection. consists four core steps: image preprocessing, feature extraction fusion, selection, classification. A novel contrast enhancement technique based on luminance information. After that, two pre-trained deep models, DarkNet-53 DensNet-201, modified in terms residual block at end trained through transfer learning. learning process, Genetic algorithm applied select hyperparameters. resultant features fused using two-step approach named serial-harmonic mean. This step increases accuracy correct classification, but some information also observed. Therefore, an developed best called marine predator optimization (MPA) controlled Reyni Entropy. selected finally classified machine classifiers final Two datasets, ISIC2018 ISIC2019, have been experimental process. On these obtained maximum 85.4% 98.80%, respectively. To prove effectiveness methods, detailed comparison conducted several recent techniques shows framework outperforms.

Language: Английский

---

Explainable deep inherent learning for multi-classes skin lesion classification

Applied Soft Computing, Journal Year: 2024, Volume and Issue: 159, P. 111624 - 111624

Published: April 19, 2024

Language: Английский

---

DVFNet: A deep feature fusion-based model for the multiclassification of skin cancer utilizing dermoscopy images

PLoS ONE, Journal Year: 2024, Volume and Issue: 19(3), P. e0297667 - e0297667

Published: March 20, 2024

Skin cancer is a common affecting millions of people annually. cells inside the body that grow in unusual patterns are sign this invasive disease. The then spread to other organs and tissues through lymph nodes destroy them. Lifestyle changes increased solar exposure contribute rise incidence skin cancer. Early identification staging essential due high mortality rate associated with In study, we presented deep learning-based method named DVFNet for detection from dermoscopy images. To detect images pre-processed using anisotropic diffusion methods remove artifacts noise which enhances quality A combination VGG19 architecture Histogram Oriented Gradients (HOG) used research discriminative feature extraction. SMOTE Tomek resolve problem imbalanced multiple classes publicly available ISIC 2019 dataset. This study utilizes segmentation pinpoint areas significantly damaged cells. vector map created by combining features HOG VGG19. Multiclassification accomplished CNN maps. achieves an accuracy 98.32% on Analysis variance (ANOVA) statistical test validate model's accuracy. Healthcare experts utilize model at early clinical stage.

Language: Английский

---

Skin Cancer Segmentation and Classification Using Vision Transformer for Automatic Analysis in Dermatoscopy-Based Noninvasive Digital System

International Journal of Biomedical Imaging, Journal Year: 2024, Volume and Issue: 2024, P. 1 - 18

Published: Feb. 3, 2024

Skin cancer is a significant health concern worldwide, and early accurate diagnosis plays crucial role in improving patient outcomes. In recent years, deep learning models have shown remarkable success various computer vision tasks, including image classification. this research study, we introduce an approach for skin classification using transformer, state-of-the-art architecture that has demonstrated exceptional performance diverse analysis tasks. The study utilizes the HAM10000 dataset; publicly available dataset comprising 10,015 lesion images classified into two categories: benign (6705 images) malignant (3310 images). This consists of high-resolution captured dermatoscopes carefully annotated by expert dermatologists. Preprocessing techniques, such as normalization augmentation, are applied to enhance robustness generalization model. transformer adapted task. model leverages self-attention mechanism capture intricate spatial dependencies long-range within images, enabling it effectively learn relevant features Segment Anything Model (SAM) employed segment cancerous areas from images; achieving IOU 96.01% Dice coefficient 98.14% then pretrained used architecture. Extensive experiments evaluations conducted assess our approach. results demonstrate superiority over traditional architectures general with some exceptions. Upon experimenting on six different models, ViT-Google, ViT-MAE, ViT-ResNet50, ViT-VAN, ViT-BEiT, ViT-DiT, found out ML achieves 96.15% accuracy Google’s ViT patch-32 low false negative ratio test dataset, showcasing its potential effective tool aiding dermatologists cancer.

Language: Английский

---

Skin Cancer Detection Using Transfer Learning and Deep Attention Mechanisms

Diagnostics, Journal Year: 2025, Volume and Issue: 15(1), P. 99 - 99

Published: Jan. 3, 2025

Background/Objectives: Early and accurate diagnosis of skin cancer improves survival rates; however, dermatologists often struggle with lesion detection due to similar pigmentation. Deep learning transfer models have shown promise in diagnosing cancers through image processing. Integrating attention mechanisms (AMs) deep has further enhanced the accuracy medical classification. While significant progress been made, research is needed improve accuracy. Previous studies not explored integration pre-trained Xception model for binary classification cancer. This study aims investigate impact various on model’s performance detecting benign malignant lesions. Methods: We conducted four experiments HAM10000 dataset. Three integrated self-attention (SL), hard (HD), soft (SF) mechanisms, while fourth used standard without mechanisms. Each mechanism analyzed features from uniquely: examined input relationships, hard-attention selected elements sparsely, soft-attention distributed focus probabilistically. Results: AMs into architecture effectively its performance. The alone was 91.05%. With AMs, increased 94.11% using self-attention, 93.29% attention, 92.97% attention. Moreover, proposed outperformed previous terms recall metrics, which are crucial investigations. Conclusions: These findings suggest that can enhance relation complex imaging tasks, potentially supporting earlier improving treatment outcomes.

Language: Английский

---

SkinNet-INIO: Multiclass Skin Lesion Localization and Classification Using Fusion-Assisted Deep Neural Networks and Improved Nature-Inspired Optimization Algorithm

Diagnostics, Journal Year: 2023, Volume and Issue: 13(18), P. 2869 - 2869

Published: Sept. 6, 2023

Background: Using artificial intelligence (AI) with the concept of a deep learning-based automated computer-aided diagnosis (CAD) system has shown improved performance for skin lesion classification. Although convolutional neural networks (DCNNs) have significantly many image classification tasks, it is still difficult to accurately classify lesions because lack training data, inter-class similarity, intra-class variation, and inability concentrate on semantically significant parts. Innovations: To address these issues, we proposed an learning best feature selection framework multiclass in dermoscopy images. The performs preprocessing step at initial contrast enhancement using new technique that based dark channel haze top–bottom filtering. Three pre-trained models are fine-tuned next trained transfer concept. In fine-tuning process, added removed few additional layers lessen parameters later selected hyperparameters genetic algorithm (GA) instead manual assignment. purpose hyperparameter GA improve performance. After that, deeper layer each network features extracted. extracted fused novel serial correlation-based approach. This reduces vector length serial-based approach, but there little redundant information. We anti-Lion optimization this issue. finally classified machine algorithms. Main Results: experimental process was conducted two publicly available datasets, ISIC2018 ISIC2019. Employing obtained accuracy 96.1 99.9%, respectively. Comparison also state-of-the-art techniques shows accuracy. Conclusions: successfully enhances cancer region. Moreover, framework. fusion version maintains shorten computational time.

Language: Английский

---

An effective multiclass skin cancer classification approach based on deep convolutional neural network

Cluster Computing, Journal Year: 2024, Volume and Issue: unknown

Published: June 17, 2024

Abstract Skin cancer is one of the most dangerous types due to its immediate appearance and possibility rapid spread. It arises from uncontrollably growing cells, rapidly dividing cells in area body, invading other bodily tissues, spreading throughout body. Early detection helps prevent progress reaching critical levels, reducing risk complications need for more aggressive treatment options. Convolutional neural networks (CNNs) revolutionize skin diagnosis by extracting intricate features images, enabling an accurate classification lesions. Their role extends early detection, providing a powerful tool dermatologists identify abnormalities their nascent stages, ultimately improving patient outcomes. This study proposes novel deep convolutional network (DCNN) approach classifying The proposed DCNN model evaluated using two unbalanced datasets, namely HAM10000 ISIC-2019. compared with transfer learning models, including VGG16, VGG19, DenseNet121, DenseNet201, MobileNetV2. Its performance assessed four widely used evaluation metrics: accuracy, recall, precision, F1-score, specificity, AUC. experimental results demonstrate that outperforms (DL) models utilized these datasets. achieved highest accuracy ISIC-2019 $$98.5\%$$ 98.5 % $$97.1\%$$ 97.1 , respectively. These show how competitive successful overcoming problems caused class imbalance raising accuracy. Furthermore, demonstrates superior performance, particularly excelling terms recent studies utilize same which highlights robustness effectiveness DCNN.

Language: Английский

---

Guaranteeing Correctness in Black-Box Machine Learning: A Fusion of Explainable AI and Formal Methods for Healthcare Decision-Making

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 90299 - 90316

Published: Jan. 1, 2024

In recent years, Explainable Artificial Intelligence (XAI) has attracted considerable attention from the research community, primarily focusing on elucidating opaque decision-making processes inherent in complex black-box machine learning systems such as deep neural networks. This spike interest originates widespread adoption of models, particularly critical domains like healthcare and fraud detection, highlighting pressing need to understand validate their mechanisms rigorously. addition, prominent XAI techniques, including LIME (Local Interpretable Model-Agnostic Explanations) SHAP (Shapley Additive exPlanations), rely heuristics cannot guarantee correctness explanations provided. article systematically addresses this issue associated with underscoring XAI's pivotal role promoting model transparency enhance quality. Furthermore, study advocates integrating Formal Methods provide guarantees for internal decision-making. The proposed methodology unfolds three stages: firstly, training models using networks generate synthetic datasets; secondly, employing techniques interpret visualize processes; finally, decision trees datasets implement ensuring model's To approach, experimentation was conducted four widely recognized medical datasets, Wisconsin Breast Cancer Thyroid (TC) which are available UCI Machine Learning Repository. Specifically, represents a significant contribution by pioneering novel approach that seamlessly integrates Methods, thereby furnishing within domain.

Language: Английский

---

Grid-Based Structural and Dimensional Skin Cancer Classification with Self-Featured Optimized Explainable Deep Convolutional Neural Networks

International Journal of Molecular Sciences, Journal Year: 2024, Volume and Issue: 25(3), P. 1546 - 1546

Published: Jan. 26, 2024

Skin cancer is a severe and potentially lethal disease, early detection critical for successful treatment. Traditional procedures diagnosing skin are expensive, time-intensive, necessitate the expertise of medical practitioner. In recent years, many researchers have developed artificial intelligence (AI) tools, including shallow deep machine learning-based approaches, to diagnose cancer. However, AI-based diagnosis faces challenges in complexity, low reproducibility, explainability. To address these problems, we propose novel Grid-Based Structural Dimensional Explainable Deep Convolutional Neural Network accurate interpretable classification. This model employs adaptive thresholding extracting region interest (ROI), using its dynamic capabilities enhance accuracy identifying cancerous regions. The VGG-16 architecture extracts hierarchical characteristics lesion images, leveraging recognized feature extraction. Our proposed leverages grid structure capture spatial relationships within lesions, while dimensional features extract relevant information from various image channels. An Adaptive Intelligent Coney Optimization (AICO) algorithm employed self-feature selected optimization fine-tuning hyperparameters, which dynamically adapts optimize extraction was trained tested ISIC dataset 10,015 dermascope images MNIST 2357 malignant benign oncological diseases. experimental results demonstrated that achieved CSI values 0.96 0.97 TP 80 dataset, 17.70% 16.49% more than lightweight CNN, 20.83% 19.59% DenseNet, 18.75% 17.53% 6.25% 6.18% Efficient Net-B0, 5.21% 5.15% over ECNN, 2.08% 2.06% COA-CAN, ARO-ECNN. Additionally, AICO ECNN exhibited minimal FPR FNR 0.03 0.02, respectively. attained loss 0.09 0.18 indicating this research outperforms existing techniques. improves accuracy, interpretability, robustness classification, ultimately aiding clinicians

Language: Английский

---