A simple linear readout of MT supports motion direction-discrimination performance

Area of Science:

• Neuroscience
• Computational Neuroscience
• Sensory Perception

Background:

• Motion discrimination relies on processing sensory signals for perceptual decisions.
• Previous studies suggested simple linear readouts for middle temporal area (MT) neuron activity in motion perception.
• Optimal decoding theoretically requires considering interneuronal correlations and individual neuron sensitivity.

Purpose of the Study:

• To determine the required sophistication of neural decoders for motion discrimination.
• To compare the performance of neurally plausible decoders against optimal and psychophysical benchmarks.
• To investigate whether correlation-blind decoders can match optimal performance in motion discrimination.

Main Methods:

• Trained various neurally plausible decoders to discriminate motion direction using simultaneous recordings from MT neuron ensembles.
• Evaluated decoder performance with and without knowledge of interneuronal correlations.
• Compared decoder performance to theoretical optimal decoders and animal psychophysical performance.

Main Results:

• Decoding motion direction without considering interneuronal correlations matched the performance of an optimal, correlation-aware decoder.
• Simple decoders achieved psychophysical performance levels with moderate temporal integration (up to 50% of stimulus duration).
• Decoders inherited complex temporal dynamics directly from the time-varying MT neural responses.

Conclusions:

• Simple, linear decoders operating on small MT neuron ensembles are sufficient for matching both optimal sensitivity and psychophysical performance.
• Correlation-blind decoding strategies can achieve near-optimal performance in motion discrimination tasks.
• The temporal dynamics of perceptual decisions can be directly inherited from sensory processing without explicit complex temporal computations.