DeDA: Deep Directed Accumulator

Hang Zhang, Rongguang Wang, Renjiu Hu, Jinwei Zhang, Jiahao Li

Abstract: Chronic active multiple sclerosis lesions, also referred to as rim+ lesions, are characterized by a hyperintense rim observed at the lesion’s edge on quantitative susceptibility maps. Despite their geometrically simple structure, characterized by radially oriented gradients at the lesion edge with a greater gradient magnitude compared to non-rim+ (rim-) lesions, recent studies indicate that the identification performance for these lesions is subpar due to limited data and significant class imbalance. In this paper, we propose a simple yet effective image processing operation, deep directed accumulator (DeDA), which provides a new perspective for injecting domain-specific inductive biases (priors) into neural networks for rim+ lesion identification. Given a feature map and a set of sampling grids, DeDA creates and quantizes an accumulator space into finite intervals and accumulates corresponding feature values. This DeDA operation can be regarded as a symmetric operation to the grid sampling within the forward-backward neural network framework, the process of which is order-agnostic, and can be efficiently implemented with the native CUDA programming. Experimental results on a dataset with 177 rim+ and 3986 rim- lesions show that $$10.1\%$$ of improvement in a partial (false positive rate $$<0.1$$ ) area under the receiver operating characteristic curve (pROC AUC) and $$10.2\%$$ of improvement in an area under the precision recall curve (PR AUC) can be achieved respectively comparing to other state-of-the-art methods. The source code is available online at https://github.com/tinymilky/DeDA .