A novel Deeplabv3+ and vision-based transformer model for segmentation and classification of skin lesions

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 92, P. 106084 - 106084

Published: Feb. 14, 2024

Language: Английский

---

Recent Advancements and Perspectives in the Diagnosis of Skin Diseases Using Machine Learning and Deep Learning: A Review

Diagnostics, Journal Year: 2023, Volume and Issue: 13(23), P. 3506 - 3506

Published: Nov. 22, 2023

Objective: Skin diseases constitute a widespread health concern, and the application of machine learning deep algorithms has been instrumental in improving diagnostic accuracy treatment effectiveness. This paper aims to provide comprehensive review existing research on utilization field skin disease diagnosis, with particular focus recent widely used methods learning. The present challenges constraints were also analyzed possible solutions proposed. Methods: We collected works from literature, sourced distinguished databases including IEEE, Springer, Web Science, PubMed, emphasis most 5-year advancements. From extensive corpus available research, twenty-nine articles relevant segmentation dermatological images forty-five about classification incorporated into this review. These systematically categorized two classes based computational utilized: traditional algorithms. An in-depth comparative analysis was carried out, employed methodologies their corresponding outcomes. Conclusions: Present outcomes highlight enhanced effectiveness over techniques diagnosis. Nevertheless, there remains significant scope for improvement, especially associated availability diverse datasets, generalizability models, interpretability models continue be pressing issues. Moreover, future should appropriately shifted. A amount is primarily focused melanoma, consequently need broaden pigmented dermatology future. insights not only emphasize potential diagnosis but directions that on.

Language: Английский

---

SkinNet-8: An Efficient CNN Architecture for Classifying Skin Cancer on an Imbalanced Dataset

Published: Feb. 23, 2023

Skin cancer is a fatal disease that has become the leading cause of death worldwide in recent years, although it curable if diagnosed early. Early skin detection significantly improves patients' chances survival and reduces mortality. In this research, we conduct experiments on high imbalance dermoscopic ISIC 2020 dataset. The primary objective study to develop shallow CNN architecture complete classification task effectively, requiring fewer computational resources without compromising accuracy. We have used pre-processing techniques remove image noise truncation augmentation balance dataset before fitting into model. Multiple performance measurement metrics were utilized establish overall performance. Our proposed model yields remarkable test accuracy 98.81%. compare our models' with different transfer learning (TL) models assess faster convergence rate. demonstrated its robustness by outperforming other TL terms within short processing time. It reasonable assume system will reliably aid dermatologists diagnosing patients early increasing rates.

Language: Английский

---

Combining State-of-the-Art Pre-Trained Deep Learning Models: A Noble Approach for Skin Cancer Detection Using Max Voting Ensemble

Diagnostics, Journal Year: 2023, Volume and Issue: 14(1), P. 89 - 89

Published: Dec. 30, 2023

Skin cancer poses a significant healthcare challenge, requiring precise and prompt diagnosis for effective treatment. While recent advances in deep learning have dramatically improved medical image analysis, including skin classification, ensemble methods offer pathway further enhancing diagnostic accuracy. This study introduces cutting-edge approach employing the Max Voting Ensemble Technique robust classification on ISIC 2018: Task 1-2 dataset. We incorporate range of cutting-edge, pre-trained neural networks, MobileNetV2, AlexNet, VGG16, ResNet50, DenseNet201, DenseNet121, InceptionV3, ResNet50V2, InceptionResNetV2, Xception. These models been extensively trained datasets, achieving individual accuracies ranging from 77.20% to 91.90%. Our method leverages synergistic capabilities these by combining their complementary features elevate performance further. In our approach, input images undergo preprocessing model compatibility. The integrates with architectures weights preserved. For each lesion under examination, every produces prediction. are subsequently aggregated using max voting technique yield final majority-voted class serving as conclusive Through comprehensive testing diverse dataset, outperformed models, attaining an accuracy 93.18% AUC score 0.9320, thus demonstrating superior reliability evaluated effectiveness proposed HAM10000 dataset ensure its generalizability. delivers robust, reliable, tool cancer. By utilizing power advanced we aim assist professionals timely accurate diagnoses, ultimately reducing mortality rates patient outcomes.

Language: Английский

---

Artificial Intelligence Applied to Non-Invasive Imaging Modalities in Identification of Nonmelanoma Skin Cancer: A Systematic Review

Cancers, Journal Year: 2024, Volume and Issue: 16(3), P. 629 - 629

Published: Feb. 1, 2024

The objective of this study is to systematically analyze the current state literature regarding novel artificial intelligence (AI) machine learning models utilized in non-invasive imaging for early detection nonmelanoma skin cancers. Furthermore, we aimed assess their potential clinical relevance by evaluating accuracy, sensitivity, and specificity each algorithm assessing risk bias. Two reviewers screened MEDLINE, Cochrane, PubMed, Embase databases peer-reviewed studies that focused on AI-based cancer classification involving cancers were published between 2018 2023. search terms included neoplasms, nonmelanoma, basal-cell carcinoma, squamous-cell diagnostic techniques procedures, intelligence, algorithms, computer systems, dermoscopy, reflectance confocal microscopy, optical coherence tomography. Based results, only directly answered review objectives efficacy measures recorded. A QUADAS-2 assessment bias was then conducted. total 44 our review; 40 utilizing 3 using microscopy (RCM), 1 hyperspectral epidermal (HEI). average accuracy AI algorithms applied all modalities combined 86.80%, with same dermoscopy. Only one three applying RCM measured a result 87%. Accuracy not regard based HEI interpretation. exhibited an overall favorable performance diagnosis via noninvasive techniques. Ultimately, further research needed isolate pooled as many testing datasets also include melanoma other pigmented lesions.

Language: Английский

---