Related Experiment Video
Updated: Jun 14, 2025
09:37
Combining Reflectance Confocal Microscopy with Optical Coherence Tomography for Noninvasive Diagnosis of Skin Cancers via Image Acquisition
Published on: August 18, 2022
2.3K
Enhancing Dermatological Diagnostics with EfficientNet: A Deep Learning Approach
Ionela Manole1, Alexandra-Irina Butacu1,2, Raluca Nicoleta Bejan3
12nd Department of Dermatology, Colentina Clinical Hospital, 020125 Bucharest, Romania.
Bioengineering (Basel, Switzerland)
|August 29, 2024
Summary
A new deep learning model for skin lesion classification achieves high accuracy, offering a faster and more cost-effective computer-aided diagnosis tool for dermatology. This advancement supports precision medicine through improved medical technology.
Area of Science:
- Dermatology
- Artificial Intelligence
- Medical Imaging
Background:
- Medical technology continues to advance, with precision medicine emerging as a key area of growth.
- Machine learning (ML) and deep learning (DL) are driving these advancements, fueled by increased computational power.
- This study focuses on a DL application for computer-aided diagnosis (CADx) in dermatology.
Purpose of the Study:
- To develop and evaluate a novel deep learning model for skin lesion classification.
- To achieve superior diagnostic performance with smaller, faster, and more cost-effective inference times compared to existing models.
- To explore the potential of DL in enhancing dermatological diagnostic capabilities.
Main Methods:
- A custom deep learning model was developed using the EfficientNetB3 architecture.
- The model was trained and validated on a dataset of 8222 skin images, including data from the authors' collection and the ISIC 2019 archive.
- The dataset encompassed six common dermatological conditions.
Main Results:
- The model achieved 95.4% validation accuracy for four major categories (melanoma, basal cell carcinoma, benign keratosis-like lesions, melanocytic nevi).
- When including two additional categories (squamous cell carcinoma, actinic keratoses) with fewer images, validation accuracy was 88.8%.
- The model demonstrated consistent accuracy on new clinical test images.
Conclusions:
- The custom deep learning model shows excellent performance in classifying diverse skin lesions.
- The model offers a promising tool for computer-aided diagnosis in dermatology.
- There is significant potential for further enhancements and applications of this DL approach.
