An overview of the role of artificial intelligence in diagnostic imaging of the nervous system

BACKGROUND: Artificial Intelligence (AI) refers to the capacity of machines, particularly computer systems, to perform tasks that typically require human intelligence. Key components of AI include Machine Learning (ML), which enables systems to learn from data and enhance their performance over time without explicit programming, and Deep Learning, a more advanced form of machine learning that utilizes multi-layered neural networks to process data. Deep learning is significantly advancing diagnostic imaging of the nervous system, particularly in neuroimaging techniques. OBJECTIVE: AI algorithms, particularly deep learning models, are utilized to analyze neuroimaging data from modalities such as MRI and CT scans. These models can swiftly and accurately detect and classify neurological abnormalities, including tumors, lesions, and signs of neurodegeneration. For instance, AI can differentiate between various types of brain tumors, aiding in precise diagnoses and efficient treatment planning. METHODS: In this review article, data was collected from multiple international scientific databases, including Google Scholar, Scopus, PubMed, and Elsevier, along with library searches in various sources. The search employed keywords like “Artificial intelligence”, “machine learning”, “deep machine learning”, “nervous system”, and medical imaging modalities such as “radiology”, “computed tomography (CT scan)”, and “magnetic resonance imaging (MRI)”. CONCLUSION: This study aims to illustrate the feasibility of automatic segmentation through artificial intelligence (AI) and the effectiveness of AI-assisted segmentation. It used MRI images, including T۲WI, T۱WI, and CE-T۱WI, of brain tumors from ۵۷ WAG/Rij rats at KU Leuven and ۴۶ mice from the cancer imaging archive (TCIA). A ۳D U-Net architecture was utilized to segment the brains affected by tumors as well as the tumors themselves. After training, the models were tested on both datasets after incorporating Gaussian noise. The study additionally examined the decrease in inter-observer variability facilitated by AI-assisted segmentation. The AI model successfully segmented tumor-affected brains from both the Leuven and TCIA datasets, attaining Dice similarity coefficients (DSCs) of ۰.۸۷ and ۰.۸۵, respectively. Following the introduction of noise, the model's performance stayed consistent as long as the signal-to-noise ratio (SNR) exceeded two or eight. For tumor lesion segmentation, the AI model yielded DSCs of ۰.۷۰ and ۰.۶۱ for the Leuven and TCIA datasets, respectively.

کلیدواژه ها:

Artificial intelligence ، machine learning ، deep machine learning ، nervous system ، radiology ، computed tomography (CT scan) ، magnetic resonance imaging (MRI)