Extending the Relative Seriality Formalism for Interpretable Deep Learning of Normal Tissue Complication Probability Models

TL;DR

This paper demonstrates that the relative seriality model can be represented as a convolutional neural network, enabling interpretable deep learning for predicting normal tissue complication probabilities in radiobiology.

Contribution

It extends the relative seriality formalism by mapping it onto CNNs, providing a biologically interpretable deep learning framework for radiobiological modeling.

Findings

Exact mapping of the seriality model onto CNN architecture

Interpretation of network layers as biological effects and tissue organization

Proof-of-principle for interpretable deep learning in radiobiology

Abstract

We formally demonstrate that the relative seriality model of Kallman, et al. maps exactly onto a simple type of convolutional neural network. This approach leads to a natural interpretation of feedforward connections in the convolutional layer and stacked intermediate pooling layers in terms of bystander effects and hierarchical tissue organization, respectively. These results serve as proof-of-principle for radiobiologically interpretable deep learning of normal tissue complication probability using large-scale imaging and dosimetry datasets.