ABSTRACT
The radiological search model (RSM) is a statistical parameterization of the Nodine-Kundel model. It consists of a search stage, characterized by two parameters: lambda and nu. The lambda parameter is the mean of a Poisson distribution describing the distribution of the numbers of non-lesion localizations per case, with smaller values indicative of avoiding generating non-lesion localizations. The second parameter is the success probability of a binomial distribution, describing the probability that a true lesion will be found by the search mechanism, with values approaching unity indicative of the ability to find true lesions. The third parameter, mu, is the separation of two unit variance normal distributions, describing the distribution of sampled decision variables from non-lesion and true lesion sites, with larger values indicative of correct classifications of found suspicious regions. The total number of RSM parameters, namely three, is one more than conventional ROC models. This is because describing search requires two parameters and describing lesion-classification performance requires one parameter. A reparameterization is described that converts the mu-dependent lambda and nu parameters to intrinsic parameters, which are independent of the mu parameter. Online appendices demonstrate Poisson and binomial sampling.
