Quiroga, R. Q., Reddy, L., Kreiman, G., Koch, C., Fried, I. (2005). Invariant visual representation by single neurons in the human brain, Nature, Vol. 435, pp. 1102-7. Sander Vaus 22.04.2015
The study Recognizing a familiar person or an object takes a fraction of a second This constitutes high-level representation How does this happen? Possibily via a subset of highly selective MTL neurons, suggesting an invariant, sparse and explicit encoding of high-level representations
The method 21 recording sessions with 8 patients suffering from pharmacologically intractable epilepsy Electrodes (with 9 microwires at their ends, 8 for recording and 1 for reference) in hippocampus, amygdala, entorhinal cortex, and parahippocampal gyrus Signal amplification using a 64-channel Neuralynx system, filtered between 1 and 9000 Hz Signals sampled at 28 khz
The method Subjects were shown images in a random order, each image 6 times for a duration of 1 second Face detection task to hold subjects attention All pictures that elicited a response in the screening session were included in the later testing sessions 3-8 different views of 7-23 different individuals or objects were used in the testing sessions (mean of 88.6 images per session)
The method The response to a picture was defined as the median number of spikes across trials between 300 and 1000 ms after stimulus onset 2 criteria for unit-responsivity: (1) The median number of spikes was larger than the average number of spikes for the baseline plus 5 standard deviation (2) The median number of spikes was at least 2
The method Distinction between single unit and multi-unit: (1) The spike shape and its variance (2) The ratio between the spike peak value and the noise level (3) The inter-spike interval distribution of each cluster (4) The presence of a refractory period for the single units (less than 1% of spikes within less than 3ms inter-spike interval) A unit s response to a stimulus was followed by a ROC analysis
The results Out of the 993 units (343 single units, 650 multiunits) recorded, 132 (64 single, 68 multi) showed a statistically significant to at least one picture All responses were highly selective: an average of only 2.8% of the presented pictures showed significant activation Out of the 132 selective units 51 showed invariance to a particular individual (38), landmark building (6), animal species (5), or object (2) 8 of the selective units showed a selective response to an individual and its name
Conclusions The results seem to indicate an abstract representation of the identity of the individual or object in the selective neurons This is not incompatible with what we know about the MTL, namely: The known anatomical connections between the higher stages of the ventral visual pathway and the MTL the well-characterized reactivity of the cortical stages feeding into the MTL to the sight of faces, objects, or spatial scenes the observation that any visual percept that will be consciously remembered later on will have to be represented in the hippocampal system
Hypotheses In the explicit (grandmoter) representations vs distributed representation debate, the authors favor a weaker version of the former The argument: (1) With distributed representation, recognition would require the simultaneous activation of a large number of cells (2) From (1) we would expect each cell to respond to many pictures with similar basic features (3) The studied MTL cells didn t respond to the great majority of images seen by the patient (4) Distributed representationalism fails (from (2) and (3)
Hypotheses This is not to imply the existence of single neurons uniquely coding for discrete percepts, because: Some responsive units responded to pictures of more than one individual or object They could only explore a tiny space of space due to time constraints That they could find in such a short time images that drive the cells suggests that each cell might represent more than one class of images