Fisher information provides a bound on the variance of any unbiased estimate for estimation tasks involving nonrandom parameters. In addition, a Fisher information approximation for ideal-observer detectability has been derived. We adopt and generalize such an approximation to establish a method to assess a system's ability to detect small changes in lesion characteristics. By representing the lesion by a size parameter, the ability to detect small changes can be approximated by a function involving the size difference and the Fisher information. A concept, termed the approximated least required difference (ALRD), is introduced and evaluated as an upper bound for assessing a system's power in size discrimination. We present a simulation study for lung nodules as an example to illustrate such a framework, where the image model incorporates a simulated CT imaging system, a thorax background and parameterized nodules. The noise is assumed to be multivariate Gaussian and the noise power spectrum (NPS) method is used to estimate the covariance matrix for the Fisher information calculation. In addition to bounding performance, our results also provide insights into factors, including nodule characteristics and acquisition parameters, that influence ALRD performance. This framework can be extended to connect other discrimination and estimation tasks, facilitating objective assessment and optimization of quantitative imaging systems.