Signal detection analysis of blindsight in monkeys

Abstract

Macaque monkeys with a unilateral lesion in V1 have been used as an animal model of blindsight. While objective proof of blindsight requires that the two aspects of blindsight (residual forced-choice localization and attenuated yes-no detection) should be tested under identical stimulus conditions using bias-free measures of sensitivity, these have not been attained in studies of nonhuman primates. Here we tested two macaque monkeys with a unilateral V1 lesion with two saccade tasks using identical stimuli: a forced-choice (FC) task and a yes-no (YN) task. An analysis based on signal detection theory revealed that sensitivity in the FC task was significantly higher than that in the YN task. Such dissociation of sensitivity between the two tasks was not observed when near-threshold visual stimuli were presented in the normal, unaffected hemifield. These results suggest that the V1-lesioned monkeys resemble the well-studied blindsight patient G.Y., whose visual experience per se was completely abolished.

Figure 1. Extent of lesion.

Left: The retinotopic map in V1 drawn on a macaque brain based on published literature. Middle and right: 3D images of the brain after the V1 lesion for two monkeys were reconstructed from the MR images. The lesion site, estimated from the MR images, is drawn in red. The dotted lines denote the border between V1 and V2. Top, medial view. Bottom, lateral view. VM, vertical meridian; HM, horizontal meridian.

Figure 2. Behavioural tasks.

Top, forced-choice (FC) task. After a fixation period, a saccadic target appeared in one of the two positions in one of the hemifield. Trials in which stimuli were presented in different hemifields were in different blocks. Bottom, Yes-no (YN) task. The ST+ condition (“YN ST+”) was identical to the FC task. Trials without a saccadic target (ST− condition) were randomly interleaved. The ratio for ST+ condition (p %) was fixed as 30% in the results for Fig. 3 and varied across trial blocks in the results for Fig. 4.

Figure 3. Performance of FC task and YN task.

The bars with black edges indicate the ratio of correct, rewarded trials to the total trials. See the text for classification of trials according to the behavioural responses. FC, FC task. YN ST+, ST+ condition of YN task. YN ST−, ST− condition of YN task. Left panel, trial blocks in which supra-threshold stimuli (target contrast 0.7) were presented in the normal hemifield (“Normal, supra-threshold”). Middle panel, trial blocks in which near-threshold stimuli (target contrast 0.05) were presented in the normal hemifield (“Normal, near-threshold”). Right panel, trial blocks in which stimuli (target contrast 0.7) were presented in the affected hemifield (“Affected”). Probability of ST+ trials (“p” in Fig. 2) was fixed at 30%. Top, monkey T. Bottom, monkey A. Dotted lines: Performance of the YN ST+ condition (P_YN) expected from that of the FC condition (P_FC): P_YN = P_FC * 2–1.

Figure 4. Behavior and analysis based on signal detection theory.

a An example of a session in which the probability of ST+ trials were varied in different blocks (on average, 161.1 trials per blocks). The value for each point represents a running average of the adjacent nine trials. Red, hit rate. Blue, false alarm rate. “FC = 1” indicates a trial block for the FC task. Monkey T, 18 months after V1 lesion. b Empirical ROC curves for Normal, near-threshold condition (left) and for Affected condition (right). Magenta, data for YN task. Green, data for FC task. A circle indicates data point for each ST+ probability. Lines indicate fitted lines. For YN task (magenta lines), the fitting was based on the unequal-variance model. Monkey T. c Sensitivity, as expressed in Da, was plotted for FC and YN task and for two trial conditions and for monkey T (top) and monkey A (bottom). ***, p < 0.001; n.s., not significant.

Figure 5. Decision bias in the YN task.

a The decision criterion for the YN task (with reference to the noise level in signal detection theory) was plotted across the ST+ probability. In all cases, the correlation coefficients were smaller than −0.81. The error bars indicate the standard deviation. b The log likelihood ratio at the criterion with 50% ST+ probability was plotted for both monkeys and two stimulus conditions. The value zero indicates optimal decision and the negative values indicate bias toward saccades over fixation. Error bars indicate the standard deviation.

Figure 6. Saccadic reaction time.

The distribution of saccadic reaction times for the two tasks and for the two stimulus conditions. See Fig. 2 for classification of response types. Monkey T. 10 ms bins.