Multilayer Perceptron Neural Network Model: A Novel Approach for LFP Contrast Sensitivity Tuning

Sahar Maleki, Reza Lashgari, Mahdi Aliyari Shoorehdeli, Mohammad Komareji · 2025

Read Multilayer Perceptron Neural Network Model: A Novel Approach for LFP Contrast Sensitivity Tuning by Sahar Maleki, Reza Lashgari, Mahdi Aliyari Shoorehdeli, Mohammad Komareji (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-19636.txt.

Introduction

Sahar Maleki and three co-authors test whether a multilayer perceptron neural network fits the contrast responses of local field potentials better than the formulas neuroscientists usually use. Local field potentials record the combined activity of neuron populations in the primary visual cortex, and their tuning curves are broader than those of single spikes. The data come from primate V1 responses to luminance contrast.

The team first sorted the recorded response curves into linear, saturating, and supersaturating types using a monotonicity index. Several identification methods were then compared, including radial basis networks, fuzzy models, and the classic Naka-Rushton function. The multilayer perceptron tuned significantly more recordings successfully across all three types, suggesting it as a better standard tool.

Catalog record

CREATORSahar Maleki, Reza Lashgari, Mahdi Aliyari Shoorehdeli, Mohammad Komareji
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-19636

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Multilayer Perceptron Neural Network Model: A Novel Approach for LFP Contrast Sensitivity Tuning Sahar Maleki, Reza Lashgari, Mahdi Aliyari Shoorehdeli, Mohammad Komareji Abstract Local field potentials (LFPs) have been demonstrated to be an important measurement to study the activity of a local population of neurons. The response tunings of LFPs have been mostly reported as weaker and broader than spike tunings. Therefore, selecting optimized tuning methods is essential for appropriately evaluating the LFP responses and comparing them with neighboring spiking activity. In this paper, new models for tuning of the contrast response functions (CRFs) are proposed. To this end, luminance contrast-evoked LFP responses recorded in primate primary visual cortex (V1) are first analyzed. Then, supersaturating CRFs are distinguished from linear and saturating CRFs by using monotonicity index (MI). The supersaturated recording data are then identified through static identification methods including multilayer perceptron (MLP) neural network, radial basis function (RBF) neural network, fuzzy model, neuro-fuzzy model, and the local linear model tree (LOLIMOT) algorithm. Our results demonstrate that the MLP neural network, compared to traditional and modified hyperbolic Naka-Rushton functions, exhibits superior performance in tuning the local field potential responses to luminance contrast stimuli, resulting in successful tuning of a significantly higher number of neural recordings of all three types. These results suggest that the MLP neural network model can be used as a novel approach to measure a better fitted contrast sensitivity tuning curve of a population of neurons than other currently used models. Keywords: Local field potential, luminance contrast sensitivity, multilayer perceptron neural network model, contrast response function, LFP tuning, primary visual cortex 1 Introduction The primate brain is the most complex and mysterious biological structure in the world, and scientists still have many unanswered questions about its function. In recent decades, scientists have made extensive efforts to understand the neuronal activity. Due to the activity of a large population of neurons, the LFP signal originating from neural populations is more difficult to interpret than single spiking neurons. Hence, an important part of these efforts has been focused on characterizing the signaling properties and functional significance of local field potentials within the cerebral cortex. For improved comprehension of LFPs, it is crucial to apply an optimized neural tuning model to LFP signals and then compare these tuned LFPs with concurrently recorded single-unit spiking activity from adjacent neurons. Precisely, optimized functional models and analysis are needed to understand the interpretation of LFP signals and the neural circuits within the cortex Einevoll et al. (2013) . The optimal local field potential tuning is characterized by its high selectivity for specific features, such as orientation, direction, and luminance contrast Lashgari et al. (2012) ; Land et al. (2013) ; Li et al. (2015) ; Jansen et al. (2015) ; Jansen et al. (2019) . Among the features, contrast sensitivity is considered one of the most important. Prior studies have demonstrated a positive correlation between stimulus luminance contrast and LFP responses Henrie & Shapley (2005) ; Peirce (2007) . Nevertheless, LFP visual signals generally do not respond linearly to contrast. Instead, they often exhibit a saturated response to high-contrast stimuli. Furthermore, some LFP responses display supersaturation, decreasing at the highest contrast levels Peirce (2007) ; Sani et al. (2013) ; Hadjipapas et al. (2015) ; Lashgari et al. (2012) . Cortical neuron and LFP responses to varying levels of contrast are commonly modeled with either the traditional Naka-Rushton function or its modified form Peirce (2007) ; Lashgari et al. (2012) . The modified Naka-Rushton model represents an improvement over the traditional version. Generally, this modified function effectively tunes neural luminance contrast responses for many neurons. However, even with this function, some neuron/LFP responses cannot be or are weakly tuned Peirce (2007) . Therefore, to overcome the shortcomings of the traditional and modified Naka-Rushton functions, a more accurate and higher-performing model for describing contrast-response functions is required. Inspired by the aforementioned discussion, this paper proposes a novel approach for LFP contrast sensitivity tuning that overcomes the limitations of prior models. In this regard, single-unit spiking activity and local field potential signals, elicited by drifting sinusoidal gratings with eight luminance contrast levels (0-76\%) , were simultaneously recorded in the V1 of awake monkeys. The collected LFP signals are first analyzed. Next, the supersaturating CRFs are distinguished from linear and saturating CRFs by using monotonicity index. Afterwards, supersaturated recording data are identified utilizing a range of static identification methods, specifically MLP neural network, RBF neural network, fuzzy model, neuro-fuzzy model, and the LOLIMOT algorithm. The performance of these methods is then compared against the currently used models for all three data types. The main body of the paper prioritizes the MLP neural network approach, while supplementary details regarding the other methods are provided in the appendix. 2 Materials and Methods The following section details the methodologies and materials of the paper. 2.1 Data Collection and Analysis The data were collected as follows: monkeys ( Macaca mulatta , n=3 males) were implanted with a scleral eye coil, and the Eyelink was used to monitor eye movements. A head post was used for head fixation, and a chronic array of ultrathin electrodes protected by a recording chamber was implanted. In the third monkey, recordings were collected daily using a single FHC electrode in the primary visual cortex. Use of the macaque monkeys and the experimental protocol were approved by the Institutional Animal Care and Use Committee (IACUC) at the State University of New York, State College of Optometry and the Use Committee at the Institute for Research in Fundamental Sciences (IPM). The monkeys were housed and handled in strict accordance with the recommendations of the National Institutes of Health (NIH) protocols and health conditions were constantly monitored by the institutional veterinary doctor. Visual stimuli were generated with a Matlab psychotoolbox and presented on a monitor (Sony Electronics; refresh rate, 160 Hz; mean luminance, 61 {\rm{cd}}/{{\rm{m}}^{2}} ; resolution, 640\times 480 pixels). The response properties quantified in this study were measured with drifting gratings presented for 2 {\rm{sec}} at 2 Hz. After obtaining the feature properties of single isolated neurons; e.g., orientation and spatial frequency tunings, the stimuli were matched with the optimized response of neurons. Contrast sensitivity was measured with eight different luminance contrasts ( 0-76\% ). The monkeys were trained to touch a bar and fixate on a small cross ( 0.1^{\circ}/{\rm{side}} ) presented at a distance of 57 {\rm{cm}} . They maintained the eye position within \pm 0.5-{1^{\circ}}\; for 3 {\rm{sec}} and released the bar when the cross changed color to receive a drop of juice as reward. During the fixation, stimuli were presented at the receptive field location of the recorded single neuron to study the contrast tuning of the single neurons and LFPs. Local field potential signals and single-unit spiking activity were simultaneously recorded in the V1 of awake monkeys. The electrodes were independently moved with individual microdrives, and the electrical signals from each electrode were collected and recorded Lashgari et al. (2012) ; Fayyaz et al. (2019) ; Jansen et al. (2019) . A total of 66 neurons in V1 of three monkeys were recorded by means of an array of brain-implanted electrodes, as well as through daily recordings with a single FHC electrode.

Read the complete text