Search papers, labs, and topics across Lattice.
This paper reformulates the Bayesian Ideal Observer (IO) test statistic using a score-based approach, specifically through the introduction of a score-based ideal observer (SIO) that leverages denoising convolutional neural networks trained on signal-absent images. By doing so, the SIO circumvents the need for extensive posterior sampling and retraining for different detection tasks, allowing for efficient performance approximation across various additive signals. Numerical experiments confirm that the SIO closely approximates the performance of the traditional IO in a signal-known-exactly detection scenario, marking a significant advancement in detection task methodologies.
The score-based ideal observer can approximate Bayesian performance without the heavy computational burden of posterior sampling or task-specific retraining.
The Bayesian Ideal Observer (IO) establishes the theoretical upper bound on task performance for binary detection tasks. However, analytical computation of the IO test statistic is generally intractable. Numerical approaches based on Markov-chain Monte Carlo (MCMC) methods, including their recent deep generative model-based extensions, typically require extensive posterior sampling for each test image. Supervised learning has also been investigated to approximate the IO performance. However, such methods are typically trained for a specific detection task and signal and may require retraining when the task or signal changes. The score function, defined as the gradient of the log probability density, encodes the local geometry of the data distribution and is a fundamental quantity in modern score-based generative modeling. This work reformulates the IO test statistic in terms of the score function and introduces a score-based ideal observer (SIO). The proposed SIO uses a denoising convolutional neural network trained exclusively on signal-absent images to estimate the signal-absent score function. Once trained, the resulting score model can be used to approximate the IO test statistic for detection tasks involving arbitrary additive signals, without per-image posterior sampling or signal-specific retraining. Numerical studies consider a signal-known-exactly (SKE) detection task with a stochastic lumpy-background model. The results demonstrate that the proposed SIO can closely approximate the IO performance.
