Artificial intelligence approach to classify unipolar and bipolar depressive disorders

Abstract

Machine learning approaches for medical decision-making processes are valuable when both high classification accuracy and less feature requirements are satisfied. Artificial neural networks (ANNs) successfully meet the first goal with its adaptive engine, while natureinspired algorithms are focusing on the feature selection (FS) process in order to eliminate less informative and less discriminant features. Besides engineering applications of ANN and FS algorithms, medical informatics is another emerging field using similar methods for medical data processing. Classification of psychiatric disorders is one of the major focus of medical informatics using artificial intelligence approaches. Being one of the most debilitating psychiatric diseases, bipolar disorder (BD) is frequently misdiagnosed as unipolar disorder (UD), leading to suboptimal treatment and poor outcomes. Thus, discriminating UD and BD at earlier stages of illness could therefore help to facilitate efficient and specific treatment. The use of quantitative electroencephalography (EEG) cordance as a biomarker has greatly enhanced the clinical utility of EEG in psychiatric and neurological subjects. In this context, the paper puts forward a study using two-step hybridized methodology: particle swarm optimization (PSO) algorithm for FS process and ANN for training process. The noteworthy performance of ANN-PSO approach stated that it is possible to discriminate 31 bipolar and 58 unipolar subjects using selected features from alpha and theta frequency bands with 89.89 % overall classification accuracy.

Major Depressive Disorder Classification Based on Different Convolutional Neural Network Models: Deep Learning Approach

“…So, the use of DL approach is another valuable perspective focusing on the structure of raw data rather than a processed biomarker. [72][73][74][75] CNNs are similar to neural networks, but the most significant difference is the convolution operation and does not require a priori assumptions on the parameters for distinction. The matrix multiplication in the conventional neural network layers is done by a matrix of parameters with a separate parameter defining the reciprocal action among each input unit and each output unit.…”

Use of EEG for Predicting Treatment Response to Transcranial Magnetic Stimulation in Obsessive Compulsive Disorder

“…Additionally, each node has an externally applied bias that is set to a negative or positive value, which then increases or decreases the net input of the activation function. 22,23 ANN's are organized as layers of nodes, namely the input layer, hidden layers, and the output layer. The number of nodes in each layer, the number of hidden layers, and the number of connections between neurons (ie, the feedforward network and recurrent network) constitute the architecture of an ANN.…”

“…In a feedforward network the input values flow from the input layer to the output layer, whereas a recurrent network has ≥1 feedback loop that acts as the memory of the neural network. 22,23 The learning (training) process for an ANN involves feeding the input vectors to the network, so that the network improves its performance (ie, the correct recognition rate) by adjusting synaptic weights according to a learning algorithm. One of the most common learning algorithms is the backpropagation (BP) algorithm.…”

“…Additionally, each node has an externally applied bias that is set to a negative or positive value, which then increases or decreases the net input of the activation function. 22,23…”

“…In a feedforward network the input values flow from the input layer to the output layer, whereas a recurrent network has ≥1 feedback loop that acts as the memory of the neural network. 22,23…”