DFDL

DFDL learns class-specific dictionary features from histopathological images to enable discriminative feature discovery for image classification and disease grading.

Key Features:

• Automatic Feature Discovery: Automatically discovers discriminative features specific to particular classes in histopathological images.
• Class-Specific Representation: Learns features/dictionaries that represent samples from a target class while minimizing their capability to represent samples from other classes.
• Low Complexity: Employs a low computational complexity formulation suitable for large-scale histopathological image datasets.

Scientific Applications:

• Intraductal Breast Lesions: Differentiates benign and malignant intraductal breast lesion histopathological images.
• Mammalian Lung Images (ADL): Applied to mammalian lung images from the Animal Diagnostics Lab (ADL) at Pennsylvania State University for veterinary diagnostic analysis.
• Brain Tumor Images (TCGA): Extracts discriminative features for brain tumor classification and grading using images from The Cancer Genome Atlas (TCGA).

Methodology:

Discriminative dictionary learning with automatic feature discovery that learns class-specific representations by representing target-class samples while minimizing representation capability for other-class samples, implemented with a low-complexity formulation.