A Sense of Direction in Biomedical Neural Networks

Zewen Liu, Timothy F. Cootes

Abstract: We describe an approach to making a model be aware of not only intensity but also properties such as feature direction and scale. Such properties can be important when analysing images containing curvilinear structures such as vessels or fibres. We propose the General Multi-Angle Scale Convolution (G-MASC), whose kernels are arbitrarily rotatable and also fully differentiable. The model manages its directional detectors in sets, and supervises a set’s rotation symmetricity with a novel rotation penalty called PoRE. The algorithm works on pyramid representations to enable scale search. Direction and scale can be extracted from the output maps, encoded and analysed separately. Tests were conducted on three public datasets, MoNuSeg, DRIVE, and CHASE-DB1. Good performance is observed while the model requires $$1\%$$ or fewer parameters compared to other approaches such as U-Net.