Attention Deficit Hyperactivity Disorder (ADHD), a neurodevelopmental disorder, is gaining increasing public and academic attention. To illuminate the current state of ADHD research, this paper reviews a range of studies, including but not limited to epidemiology, pathological origins, patient challenges, and mainstream treatments. It also analyzes a series of experiments and studies that use automated tools to analyze physiological signals for ADHD detection and differentiation. The first section provides a comprehensive summary of existing research on the pathological origins and subtype differences of ADHD, covering multiple perspectives and levels. This lays a theoretical foundation for the idea of analyzing ADHD using EEG signals. The second section focuses on distinctions in EEG signals among individuals with normal neurodevelopment, ADHD populations, and its subtypes, demonstrating that deep learning can be used to train on such physiological signals for automated diagnosis. The third section highlights existing studies that leverage deep learning, outlining the potential developments in this field. The paper objectively analyzes and summarizes obstacles and issues that may be encountered on the research path of utilizing deep learning as a tool for automated ADHD diagnosis, offering recommendations for future development. Through this review, the progress and limitations in ADHD research become clear, providing insights for continued and more informed development in the field.