The Brain Activation-Based Sexual Image Classifier (BASIC): A Sensitive and Specific fMRI Activity Pattern for Sexual Image Processing

Abstract

Previous studies suggest there is a complex relationship between sexual and general affective stimulus processing, which varies across individuals and situations. We examined whether sexual and general affective processing can be distinguished at the brain level. In addition, we explored to what degree possible distinctions are generalizable across individuals and different types of sexual stimuli, and whether they are limited to the engagement of lower-level processes, such as the detection of visual features. Data on sexual images, nonsexual positive and negative images, and neutral images from Wehrum et al. (2013) (N = 100) were reanalyzed using multivariate support vector machine models to create the brain activation-based sexual image classifier (BASIC) model. This model was tested for sensitivity, specificity, and generalizability in cross-validation (N = 100) and an independent test cohort (N = 18; Kragel et al. 2019). The BASIC model showed highly accurate performance (94-100%) in classifying sexual versus neutral or nonsexual affective images in both datasets with forced choice tests. Virtual lesions and tests of individual large-scale networks (e.g., visual or attention networks) show that individual networks are neither necessary nor sufficient to classify sexual versus nonsexual stimulus processing. Thus, responses to sexual images are distributed across brain systems.

Keywords: erotic images; machine learning prediction model; multivariate analysis; neuroimaging; sexual stimuli processing; support vector machine classification.

Figure 1

Predictive weight maps of the BASIC model. These brain maps represent the contribution of each voxel for the classification between sexual and neutral/affective conditions. The color bar thus represents the predictive weight value. The MNI-space anatomical underlay is adapted from Keuken et al. (2014). Upper left image represents the right hemisphere. The image next to that represents the left hemisphere.

Figure 2

Cosine similarity between the BASIC and all conditions of Study 1 (A) and Study 2 (B). Lines connect data points from individual participants. The threshold calculated with optimal balanced error rate was 0.25 for Study 1 and 0.06 for Study 2. ^** = P < 0.01, ^*** = q < 0.05 FDR.

Figure 3

Receiver Operating Characteristic (ROC) plot for the BASIC model performance on sexual-positive (A) and sexual-negative (B) classification of data from Studies 1 and 2, both forced choice (FC) and single interval (SI) classification methods. The numerals 1 and 2 in the legend indicate cross-validated performance in Study 1 and generalization performance in Study 2, respectively.

Figure 4

Spatial scale evaluation for classification between sexual and neutral conditions (for [sexual vs. positive] and [sexual vs. negative], see Supplementary Figure 3) from Study 1 on whole-brain, all parcels, and individual parcel levels, based on large-scale networks (Buckner, Krienen, Castellanos, Diaz, and Yeo 2011). This reveals that whole-brain models performed better than single-network models. In addition, the model based on brain-wide within-parcel (region) averages performed as well as the model based on voxel-level patterns, indicating that fine spatial scale pattern information is not needed for accurate performance.

Figure 5

Cortical network profile for BASIC model. Pattern energy in resting-state cortical networks by Schaefer et al. (2018) is distributed unevenly. Wedges represent positive (pink) and negative (purple) weights and corresponding networks are presented for the networks with the highest and lowest average weights.