Addiction is a psychological disease that leads to cognitive damage and dysfunction in the brain. Accurate addiction diagnosis methods are crucial for patients who exhibit both physical dependence and mental problems. Various addiction tests, such as urine, blood plasma, and saliva tests, are available; however, patients often attempt to conceal their addiction during diagnostic tests. Therefore, utilizing brain signals as an alternative diagnostic test for addiction holds promise. This research aims to identify distinctive features between a group of healthy mice and a group of morphine-addicted mice using machine learning algorithms. To achieve this objective, twelve male mice were subjected to electrode implantation and drug addiction in a controlled laboratory environment. EEG signals were collected from the rats' skulls before and after addiction to examine brain data. Signal preprocessing techniques including removal of city electricity noise, average removal, reduction of sampling rate, and removal of movement artifacts were applied prior to conducting windowing operations. The feature extraction process involved analyzing various aspects of the signals, including the power of frequency subbands (delta, alpha, beta, theta, and gamma), as well as nonlinear features such as Lyapunov exponent, entropy, and fractal dimension. To facilitate this analysis, the signals were transformed from the time domain to the phase space using appropriate embedding dimensions and delay times. Chaotic features like Recurrence point density, average diagonal line length, and horizontal and vertical lines were then extracted. Subsequently, the performance of KNN, SVM, and DT algorithms was eva‎luated using these extracted features. The eva‎luation results indicated that a combination of four features (gamma band power spectrum density, average diagonal line length, Higuchi fractal dimension, and laminarity) yielded acceptable resolution. Notably, the feature "average length of diagonal lines" achieved better differentiation between healthy and addicted mice when analyzed with the KNN algorithm. This feature exhibited an average accuracy of 84.61%, an average accuracy of 84.30%, and an average sensitivity of 84.93%.

امير اخوان بي تقصير , احسان روحاني

مريم ذكري , جلال ذهبي