Introduction

When you want to get your friend's attention at a distance you don't just wave your arms. You don't just shout her name, either, but you do both at the same time. Why? Because you know that your friend will be more likely to detect you if you try to alert more than one of her senses to your presence. Many areas of her brain are involved in detecting and orienting toward you. A pivotal region is located in the midbrain and comprises the deep layers of the superior colliculus.

The deep superior colliculus (DSC) is neurophysiologically and connectionally distinct from the superficial colliculus and comprises the cytoarchitectually distinct intermediate and deep layers (Sparks & Hartwich-Young, 1989). The DSC integrates multisensory input and initiates orienting movements toward the source of stimulation (Stein & Meredith, 1993). Various brain regions involved in visual, auditory, and somatosensory processing provide input to the DSC (Cadusseau & Roger, 1985; Edwards, Ginsburgh, Henkel, & Stein, 1979; Harting, Updyke, & Van Lieshout, 1992; Sparks & Hartwich-Young, 1989). Depending on species, orienting movements produced by the DSC can include saccadic eye movements, head movements, and even whole body movements (monkey: Freedman, Stanford, & Sparks, 1996; Robinson, 1972; Schiller & Stryker, 1972; Stryker & Schiller, 1975; cat: Guitton, Crommelinck, & Roucoux, 1980; Harris, 1980; Paré, Crommelinck, & Guitton, 1994; Roucoux, Guitton, & Crommelinck, 1980; rodent: Sahibzada, Dean, & Redgrave, 1986; toad: Ewert, 1970).

The DSC is organized topographically, so that neurons that are neighbors in the DSC have receptive field centers that are neighbors in the environment (auditory: Middlebrooks & Knudsen, 1984; visual: Meredith & Stein, 1990; somatosensory: Meredith, Clemo, & Stein, 1991). Sensory maps for different modalities are roughly in register in the DSC. Central to analysis of multisensory integration in the DSC is the fact that individual neurons can receive input of more than one sensory modality (Meredith & Stein, 1983, 1985, 1986b; Meredith, Wallace, & Stein, 1992; Stein & Arigbede, 1972; Stein, Magalhaes-Castro, & Kruger, 1976; Wallace, Meredith, & Stein, 1998; Wallace & Stein, 1996; Wallace, Wilkinson, & Stein, 1996). A multisensory DSC neuron has a receptive field for each sensory modality it receives, and these receptive fields overlap (Gordon, 1973; Kadunce, Vaughan, Wallace, & Stein, 2001; Meredith & Stein, 1996; Wickelgren, 1971). Thus, multisensory integration in the DSC can take place on individual neurons.

Multisensory integration in the DSC is of two basic types: enhancement and depression (King & Palmer, 1985; Meredith & Stein, 1983, 1986a, 1986b; Wallace et al., 1996, 1998). They are defined as the augmentation and diminution, respectively, of the response of a neuron to a stimulus of one modality by the presentation of a second stimulus of a different modality. Multisensory interactions depend on the spatial and temporal relationships between the stimuli (Kadunce et al., 2001; Meredith, Nemitz, & Stein, 1987; Meredith & Stein, 1986a, 1996). Stimuli that occur at the same time and place may produce enhancement, while stimuli that occur at different times or places may produce depression. Multisensory depression may result from competitive interactions occurring within the DSC (Findlay & Walker, 1999; Munoz & Istvan, 1998). Multisensory enhancement (MSE) may improve the ability of an organism to detect targets in the environment (Stein, Huneycutt, & Meredith, 1988; Stein, Meredith, Huneycutt, & McDade, 1989; Wilkinson, Meredith, & Stein, 1996). This chapter describes the analysis and modeling of MSE.

Percent MSE (%MSE) has been quantified using the following formula (Meredith & Stein, 1986b): %MSE=( CM-M S Max M S Max ) ×100% (1) where CM is the number of neural impulses evoked by cross-modal stimulation and MS_Max is the number of impulses evoked when the more effective of two modality-specific stimuli is presented alone. Percent MSE can range upward of 1000% (Meredith & Stein, 1986b), but it is dependent on the magnitudes of the modality-specific responses. By the property known as inverse effectiveness, smaller modality-specific responses are associated with larger %MSE (Meredith & Stein, 1986b; Wallace & Stein, 1994). Despite the enhanced responsiveness provided by MSE, not all DSC neurons are multisensory. About 46% of DSC neurons in cat and 73% in monkey are unimodal, receiving sensory input of only one modality (Wallace & Stein, 1996). These findings raise intriguing questions.

Neurons in the DSC do more than simply respond to the larger of their cross-modal inputs. They exhibit MSE, which produces enhanced responses that may even be larger than the sum of their responses to the two modality-specific inputs presented alone. What, then, is the functional role of MSE? If an enhanced response is important for signal processing in the DSC, then why is MSE magnitude-dependent? If MSE enhances the responsiveness of DSC neurons, then why are not all DSC neurons multisensory? This chapter introduces our analytical and modeling research, based on probability and information theory, that attempts to answer these questions. Our model, based on Bayes' rule, suggests a functional role for MSE and provides a possible explanation for inverse effectiveness (Anastasio, Patton, & Belkacem-Boussaid, 2000). An experimentally testable prediction can be derived from the Bayes' rule model, which we will illustrate. We also present an information theoretic analysis of the model that offers a possible explanation for why some DSC neurons are multisensory but others are not (Patton, Belkacem-Boussaid, & Anastasio, 2002). This modeling work opens up a new perspective on the phenomenon of MSE.