A feature fusion system for basal cell carcinoma detection through data-driven feature learning and patient profile

Basal cell carcinoma (BCC) is the most common skin cancer, which is highly damaging in its advanced stages. Computer-aided techniques provide a feasible option for early detection of BCC. However, automated BCC detection techniques immensely rely on handcrafting high-level precise features. Such features are not only computationally complex to design but can also represent a very limited aspect of the lesion characteristics. This paper proposes an automated BCC detection technique that directly learns the features from image data, eliminating the need for handcrafted feature design.

The proposed method is composed of 2 parts. First, an unsupervised feature learning framework is proposed which attempts to learn hidden characteristics of the data including vascular patterns directly from the images. This is done through the design of a sparse autoencoder (SAE). After the unsupervised learning, we treat each of the learned kernel weights of the SAE as a filter. Convolving each filter with the lesion image yields a feature map. Feature maps are condensed to reduce the dimensionality and are further integrated with patient profile information. The overall features are then fed into a softmax classifier for BCC classification.

On a set of 1199 BCC images, the proposed framework achieved an area under the curve of 91.1%, while the visualization of learned features confirmed meaningful clinical interpretation of the features.

The proposed framework provides a non-invasive fast BCC detection tool that incorporates both dermoscopic lesional features and clinical patient information, without the need for complex handcrafted feature extraction.