Robust Influence-based Training Methods for Noisy Brain MRI

TL;DR

The paper tackles robust brain tumor classification from MRI when training data are noisy. It proposes two influence-based training strategies, ISR and ISP, that leverage influence functions to reweight or perturb training samples; for ISR, the weighted loss is $L_w(D_{trn}, \theta, w)$ and sample weights derive from $I_{up, loss}$, while for ISP, perturbations follow $I_{pert, loss}$ to form a perturbed training set. On the Brain Tumor Dataset, under Gaussian and Rician noise, ISR/ISP consistently improve robustness, with ISP often achieving the best accuracy and outperforming Naive, Adversarial Training, and Noise2Void baselines. Importantly, these methods do not require denoising or clean data and can extend to other medical imaging tasks, with future work pointing toward semi-supervised/unsupervised extensions that further enhance learning from noisy medical images.

Abstract

Correctly classifying brain tumors is imperative to the prompt and accurate treatment of a patient. While several classification algorithms based on classical image processing or deep learning methods have been proposed to rapidly classify tumors in MR images, most assume the unrealistic setting of noise-free training data. In this work, we study a difficult but realistic setting of training a deep learning model on noisy MR images to classify brain tumors. We propose two training methods that are robust to noisy MRI training data, Influence-based Sample Reweighing (ISR) and Influence-based Sample Perturbation (ISP), which are based on influence functions from robust statistics. Using the influence functions, in ISR, we adaptively reweigh training examples according to how helpful/harmful they are to the training process, while in ISP, we craft and inject helpful perturbation proportional to the influence score. Both ISR and ISP harden the classification model against noisy training data without significantly affecting the generalization ability of the model on test data. We conduct empirical evaluations over a common brain tumor dataset and compare ISR and ISP to three baselines. Our empirical results show that ISR and ISP can efficiently train deep learning models robust against noisy training data.

Figures (2)

• Figure 1: Visualization of perturbation by our ISP.
• Figure 2: Training samples crafted by each training method.