The application of Spiking Neural Networks in Schizophrenia

Background and aims: Nowadays, machine learning is widely used in various medical researchprojects. Deep learning is one of the well-known subsets of machine learning, which includes differentNeural Network (NN) types. Spiking Neural Networks (SNNs) that closely mimic naturalneural networks are one of the brand-new concepts of the next generation of NNs. In addition toneuronal and synaptic status, SNNs incorporate time into their working model.Schizophrenia is a serious mental disorder in which people interpret reality abnormally. Schizophreniamay result in some combination of hallucinations, delusions, and extremely disorderedthinking and behavior that impairs daily functioning and can be disabling. One of the most discussedtheories about Schizophrenia etiology is the idea of functional disconnections. The mainmotivation is that Schizophrenia cannot be explained by an impairment of a single brain regionbut only by a decreased interaction between multiple brain regions. Our research goal is to modelSchizophrenia with SNNs based on pruning the connections between neurons of different regions.Method: In this article, the keywords “Schizophrenia,” “Spiking Neural Networks,” and “NeuralNetworks” have been searched in international databases of articles such as PubMed, GoogleScholar, Science Direct, Elsevier, Scopus, and proper articles were extracted, and reviewed. Accordingto the available articles in those databases, no article that has worked on modeling Schizophreniawith spiking neural networks was found.Results: Spiking Neural Networks are superior but more complex than traditional neural networksin many aspects. SNNs can perfectly model the malfunction of neuron connectivity (disconnection)since they emulate biological neuronal functionality by processing visual informationwith binary events (i.e., spikes) over multiple time steps. Therefore, SNNs are considered suitablemodels for processing spatiotemporal brain data (STBD) because they can be implemented usingseveral models. When an SNN is already working, it can still train. Also, to train an SNN, yousimply need to train the output neurons. SNNs typically have fewer neurons in comparison toTraditional ANNs. Last but not least, SNNs can work incredibly quickly due to sending impulsesinstead of a continuous value. According to the above-mentioned traits of SNNs and by consideringdiscussed studies for Schizophrenia, disconnection is primarily implemented by an increasedpruning of synapses. Such pruning is a normal developmental process between adolescence andearly adulthood. Computational models demonstrate that too strong pruning can cause fragmentedrecall or the recall of new patterns, which can be related to the symptom of hallucinations inSchizophrenia.Conclusion: SNNs are difficult to train due to various hyper-parameters and their training time.And so far, there is no learning algorithm built expressly for modeling Schizophrenia. Also, buildinga small SNN is impracticable. Notably, the Schizophrenia symptoms replicated with connectionpruning focus solely on hallucinations or delusions. In fact, it might be more appropriate todisturb connections between neurons in a biological context instead of simply cutting them. Inconclusion, it can be assumed that hallucinations in patients diagnosed with Schizophrenia can bemodeled by SNN methods based on disconnection to predict patients’ conditions.

کلیدواژه ها:

Schizophrenia ، Spiking Neural Networks ، Neural Networks ، Artificial Intelligence