Skin Lesion Classification Using Hybrid Deep Neural Networks (1702.08434v2)

Abstract: Skin cancer is one of the major types of cancers with an increasing incidence over the past decades. Accurately diagnosing skin lesions to discriminate between benign and malignant skin lesions is crucial to ensure appropriate patient treatment. While there are many computerised methods for skin lesion classification, convolutional neural networks (CNNs) have been shown to be superior over classical methods. In this work, we propose a fully automatic computerised method for skin lesion classification which employs optimised deep features from a number of well-established CNNs and from different abstraction levels. We use three pre-trained deep models, namely AlexNet, VGG16 and ResNet-18, as deep feature generators. The extracted features then are used to train support vector machine classifiers. In the final stage, the classifier outputs are fused to obtain a classification. Evaluated on the 150 validation images from the ISIC 2017 classification challenge, the proposed method is shown to achieve very good classification performance, yielding an area under receiver operating characteristic curve of 83.83% for melanoma classification and of 97.55% for seborrheic keratosis classification.

• The paper presents a novel hybrid method that fuses deep features from multiple CNNs with SVM classifiers for robust skin lesion classification.
• It achieves impressive results with AUC scores of 83.83% for melanoma and 97.55% for seborrheic keratosis, demonstrating high diagnostic accuracy.
• The study underscores the potential for advanced, non-invasive diagnostic tools in dermatology, paving the way for timely skin cancer detection.

Hybrid Deep Neural Networks for Skin Lesion Classification: An Analysis

The paper "Skin Lesion Classification Using Hybrid Deep Neural Networks" presents an approach to the classification of skin lesions by leveraging the capabilities of diverse convolutional neural networks (CNNs). The research underscores the criticality of accurate diagnosis in skin cancer treatment, pointing to the superior efficacy of CNNs over conventional image classification methods in achieving this objective.

The authors employ a novel hybrid method that utilizes deep features from multiple pre-trained CNN models—specifically AlexNet, VGG16, and ResNet-18. These models serve as feature generators from which support vector machine (SVM) classifiers are trained and subsequently fused to produce the final classification result. The classification task focuses on three lesion types: malignant melanoma, seborrheic keratosis, and benign nevi. Evaluations are conducted using a subset of images from the ISIC 2017 classification challenge.

The results demonstrate significant accuracy in skin lesion classification, with an area under the receiver operating characteristic curve (AUC) of 83.83% for melanoma classification and 97.55% for seborrheic keratosis classification. These outcomes suggest that the use of multiple CNN architectures, combined with SVM fusion, provides a competitive classification performance that is often on par with specialized state-of-the-art algorithms tailored for skin lesion categorization. The confluence of various abstraction levels from different CNNs ensures a robust classification framework that can accommodate the variety and complexity inherent in dermoscopic images.

Implications and Future Directions

This research has meaningful implications for both the clinical and technological aspects of dermatology and medical imaging. Clinically, the automated approach offers a non-invasive, accurate, and efficient diagnostic tool that can assist dermatologists in making more informed and faster decisions, thereby enhancing patient management and potentially improving outcomes through early intervention in melanoma cases.

From a technological standpoint, the paper illustrates the potential of transferring features learned from natural images to the medical imaging domain, implying that extensive labeled datasets specific to medical applications may not always be necessary for efficient model training. However, there remains potential for this model to be expanded. Incorporating additional advanced pre-trained models like DenseNets or employing more sophisticated data augmentation and pre-processing strategies could enhance classification accuracy and generalizability. Moreover, the approach could benefit from the integration of more comprehensive datasets to further validate the model's efficacy across diverse sub-populations and conditions.

The innovative combination of CNNs in this paper presents a promising pathway towards the development of effective computer-aided diagnostic systems in dermatology. Future research could investigate the implications of different image resolutions during training, explore alternative fusion strategies, and consider the practicality of real-time applications in clinical settings to ensure that the derived benefits translate effectively into practice.

By leveraging state-of-the-art deep learning techniques, this paper contributes to a growing body of research aimed at enhancing the accuracy and efficiency of skin lesion classification, signifying an advancement in automated medical diagnostics.