Deep Learning for Melanoma Detection: A Deep Learning Approach to Differentiating Malignant Melanoma from Benign Melanocytic Nevi DOI Open Access
Magdalini Kreouzi, Nikolaos Theodorakis, Georgios Feretzakis

et al.

Cancers, Journal Year: 2024, Volume and Issue: 17(1), P. 28 - 28

Published: Dec. 25, 2024

Background/Objectives: Melanoma, an aggressive form of skin cancer, accounts for a significant proportion skin-cancer-related deaths worldwide. Early and accurate differentiation between melanoma benign melanocytic nevi is critical improving survival rates but remains challenging because diagnostic variability. Convolutional neural networks (CNNs) have shown promise in automating detection with accuracy comparable to expert dermatologists. This study evaluates compares the performance four CNN architectures—DenseNet121, ResNet50V2, NASNetMobile, MobileNetV2—for binary classification dermoscopic images. Methods: A dataset 8825 images from DermNet was standardized divided into training (80%), validation (10%), testing (10%) subsets. Image augmentation techniques were applied enhance model generalizability. The architectures pre-trained on ImageNet customized classification. Models trained using Adam optimizer evaluated based accuracy, area under receiver operating characteristic curve (AUC-ROC), inference time, size. statistical significance differences assessed McNemar’s test. Results: DenseNet121 achieved highest (92.30%) AUC 0.951, while ResNet50V2 recorded (0.957). MobileNetV2 combined efficiency competitive performance, achieving 92.19% smallest size (9.89 MB), fastest time (23.46 ms). despite its compact size, had slower (108.67 ms), slightly lower (90.94%). Performance among models statistically (p < 0.0001). Conclusions: demonstrated superior provided most efficient solution deployment resource-constrained settings. CNNs show substantial potential clinical mobile applications.

Language: Английский

YOLOSAMIC: A Hybrid Approach to Skin Cancer Segmentation with the Segment Anything Model and YOLOv8 DOI Creative Commons
Sevda Gül, Gökçen Çetinel, Bekir Murat Aydın

et al.

Diagnostics, Journal Year: 2025, Volume and Issue: 15(4), P. 479 - 479

Published: Feb. 16, 2025

Background/Objective: The rising global incidence of skin cancer emphasizes the urgent need for reliable and accurate diagnostic tools to aid early intervention. This study introduces YOLOSAMIC (YOLO SAM in Cancer Imaging), a fully automated segmentation framework that integrates YOLOv8 lesion detection, Segment Anything Model (SAM)-Box precise segmentation. objective is develop system handles complex characteristics without requiring manual Methods: A hybrid database comprising 3463 public 765 private dermoscopy images was built enhance model generalizability. employed localize lesions through bounding box while SAM-Box refined process. trained evaluated under four scenarios assess its robustness. Additionally, an ablation examined impact grayscale conversion, image blur, pruning on performance. Results: demonstrated high accuracy, achieving Dice Jaccard scores 0.9399 0.9112 0.8990 0.8445 dataset. Conclusions: proposed provides robust, solution segmentation, eliminating annotation. Integrating enhances precision, making it valuable decision-support tool dermatologists.

Language: Английский

Citations

0
Review of Non-Invasive Imaging Technologies for Cutaneous Melanoma DOI Creative Commons
Luke Horton,

Joseph W. Fakhoury,

Rayyan Manwar

et al.

Biosensors, Journal Year: 2025, Volume and Issue: 15(5), P. 297 - 297

Published: May 7, 2025

Imaging technologies are constantly being developed to improve not only melanoma diagnosis, but also staging, treatment planning, and disease progression. We start with a description of how is characterized using histology, then continue by discussing nearly two dozen different technologies, including systems currently used in medical practice those development. For each technology, we describe its method operation, it or would be projected most commonly diagnosing managing melanoma, for current use, identify at least one manufacturer. provide table the biomarkers identified main limitations associated technology conclude offering suggestions on specific characteristics that might best enhance technology’s potential widespread clinical adoption.

Language: Английский

Citations

0
Deep Learning for Melanoma Detection: A Deep Learning Approach to Differentiating Malignant Melanoma from Benign Melanocytic Nevi DOI Open Access
Magdalini Kreouzi, Nikolaos Theodorakis, Georgios Feretzakis

et al.

Cancers, Journal Year: 2024, Volume and Issue: 17(1), P. 28 - 28

Published: Dec. 25, 2024

Background/Objectives: Melanoma, an aggressive form of skin cancer, accounts for a significant proportion skin-cancer-related deaths worldwide. Early and accurate differentiation between melanoma benign melanocytic nevi is critical improving survival rates but remains challenging because diagnostic variability. Convolutional neural networks (CNNs) have shown promise in automating detection with accuracy comparable to expert dermatologists. This study evaluates compares the performance four CNN architectures—DenseNet121, ResNet50V2, NASNetMobile, MobileNetV2—for binary classification dermoscopic images. Methods: A dataset 8825 images from DermNet was standardized divided into training (80%), validation (10%), testing (10%) subsets. Image augmentation techniques were applied enhance model generalizability. The architectures pre-trained on ImageNet customized classification. Models trained using Adam optimizer evaluated based accuracy, area under receiver operating characteristic curve (AUC-ROC), inference time, size. statistical significance differences assessed McNemar’s test. Results: DenseNet121 achieved highest (92.30%) AUC 0.951, while ResNet50V2 recorded (0.957). MobileNetV2 combined efficiency competitive performance, achieving 92.19% smallest size (9.89 MB), fastest time (23.46 ms). despite its compact size, had slower (108.67 ms), slightly lower (90.94%). Performance among models statistically (p < 0.0001). Conclusions: demonstrated superior provided most efficient solution deployment resource-constrained settings. CNNs show substantial potential clinical mobile applications.

Language: Английский

Citations

1