Related Experiment Video
Updated: Jun 2, 2026
Combining Reflectance Confocal Microscopy with Optical Coherence Tomography for Noninvasive Diagnosis of Skin Cancers via Image Acquisition
Published on: August 18, 2022
A model combining deep learning and ensemble learning for melanoma recognition via dermoscopy
1Dermatology, Heilongjiang Zhongyiyaodaxue Fushu Dier Yiyuan, Harbin, Heilongjiang, China.
Background:
Accurate identification of melanoma is key to improving prognosis, while traditional dermatoscopic recognition suffers from subjectivity and poor consistency. This study aimed to develop a deep learning-ensemble model to enhance the performance of dermatoscopic differential diagnosis between malignant melanoma and other malignant skin lesions.
Methods:
Using the ISIC-2024 and HAM10000 dermatoscopic datasets, nine convolutional neural network models (DenseNet201, EfficientNetB7, EfficientNetV2S, InceptionResNetV2, MobileNetV3Large, NASNetLarge, ResNet50V2, VGG19, and Xception) were employed for initial classification; the raw scores from each model were then used as feature vectors and input into an XGBoost ensemble model to construct the final classification model. Model performance was validated through accuracy, precision, recall, F1 Score and area under the curve (AUC).
Results:
All nine models significantly distinguished between the two classes (malignant melanoma and other malignant skin lesions; all p<0.001), with AUCs on the test dataset ranging from 0.921-0.967; the XGBoost ensemble model achieved an AUC of 1.00 on the training dataset and 0.988 on the test dataset, with Xception as the primary feature source.
Conclusion:
The integration of deep learning and ensemble learning strategies can improve dermatoscopic melanomadifferential diagnosis accuracy, provide technical support for clinically assisted recognition and possessing potential for clinical translation.
