1. Utilization of fMRI with optical amplification to diagnose attention deficit hyperactivity disorder.
- Author
-
Salah, Eman, Shokair, Mona, El-Samie, Fathi E. Abd, and Shalaby, Wafaa A.
- Abstract
Attention deficit hyperactivity disorder (ADHD) is a serious condition that may affect life and lead to significant disruption of functional and brain pathways and psychological state. It is important to find effective therapeutic strategies to overcome this disease and effective ways to treat it. It is better for patients who suffer from this disease to monitor their psychological and health conditions from an early age. As soon as they suspect the possibility of this disease, they should take the initiative to try to see a physician to diagnose the conditions, and then conduct an appropriate brain examination. Recent studies have indicated that functional magnetic resonance imaging (fMRI), which is a special type of magnetic resonance imaging, has an effective role in detecting the disease. It relies on old and familiar electronics as it uses a strong magnetic field and radio waves, and it is increasingly challenged by the push toward stronger magnetic fields and a greater number of channels, which poses major problems for it. These problems can be avoided by using optical techniques. In addition, convolution neural networks (CNNs) are mainly involved in classifying the images captured by fMRI. This approach relies on CNNs for deep feature extraction. An architectural model of a CNN based on residual learning and depth sequencing strategies is proposed in this paper. The proposal consists of 23 layers, and three different algorithms are used to improve performance. A comparison is made between them. They are adaptive momentum (ADAM), stochastic gradient ratios with momentum (SGDM), and an algorithm based on root main square error called (RMSprop), and they are applied on the fMRI dataset. Using these optimization algorithms to classify ADHD cases, it was concluded that the accuracy of ADAM is 95%, SGDM is 96.11%, and RMSprop is 97.78%. The proposed CNN achieves an accuracy of 97.7% compared to the ResNet with 95.83% and the GoogleNet with 91.67%. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF