Fuzzy Structural Broad Learning For Breast Cancer Classification

We present a novel fuzzy structural broad learning model for diffusion MRI-based breast cancer classification. We first reconstruct a group of fuzzy samples from the original uncertain breast cancer data by type-2 fuzzy subsystems. To better obtain intra-class compactness and inter-class separability, we design a generalized correntropy autoencoder with strict graph constraints to fine-tune the spatial distribution of the fuzzy samples. Third, the proposed fuzzy graph structure attention network is trained by the new representation of the fuzzy samples to jointly utilize/update the features in edges and nodes and learn structural uncertainty. The final classification is solved in a closed form by calculating the pseudoinverse matrix constructed by the refined fuzzy samples and their structural features. Extensive experiments demonstrate our method can better deal with the intrinsic uncertainties of breast lesions and extract structural information of breast cancer. The proposed method outperforms other competitive methods in terms of quantitative measures, qualitative displays, and evaluation metrics for breast cancer diagnosis.