SVD-DIP: Overcoming the Overfitting Problem in DIP-based CT Reconstruction

TL;DR

This paper introduces SVD-DIP, a novel regularization technique for deep image prior-based CT reconstruction that fine-tunes only singular values to prevent overfitting and improve stability.

Contribution

It proposes a new method that restricts DIP optimization to singular value adjustments, enhancing reconstruction stability and reducing overfitting.

Findings

Significantly improved stability in CT reconstructions.

Effective prevention of overfitting to noise.

Validated on real and medical datasets.

Abstract

The deep image prior (DIP) is a well-established unsupervised deep learning method for image reconstruction; yet it is far from being flawless. The DIP overfits to noise if not early stopped, or optimized via a regularized objective. We build on the regularized fine-tuning of a pretrained DIP, by adopting a novel strategy that restricts the learning to the adaptation of singular values. The proposed SVD-DIP uses ad hoc convolutional layers whose pretrained parameters are decomposed via the singular value decomposition. Optimizing the DIP then solely consists in the fine-tuning of the singular values, while keeping the left and right singular vectors fixed. We thoroughly validate the proposed method on real-measured $\mu$CT data of a lotus root as well as two medical datasets (LoDoPaB and Mayo). We report significantly improved stability of the DIP optimization, by overcoming the overfitting to noise.