EXPRESS: Advances in artificial intelligence for neuroimaging.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Deep Learning

Background:

• Neuroimaging applications involve multi-step pipelines crucial for diagnosing neurological disorders, tracking disease progression, and guiding treatment.
• These pipelines include image acquisition, reconstruction, enhancement, registration, segmentation, diagnosis, and prognosis.

Purpose of the Study:

• To review recent advances and emerging trends in artificial intelligence (AI) and deep learning (DL) applications across diverse neuroimaging modalities.
• To highlight how new AI approaches address long-standing challenges in neuroimaging.

Main Methods:

• Survey of recent advances in AI and DL for neuroimaging.
• Discussion of specific AI approaches including physics-informed models, self-supervised learning, graph neural networks, generative diffusion models, and data harmonization.
• Inclusion of multimodal studies integrating imaging with clinical and molecular data.

Main Results:

• AI and DL are transforming neuroimaging by improving efficiency, image quality, resolution, accuracy, and clinical utility.
• New AI methods address challenges like data scarcity (self-supervised learning), complex data structures (graph neural networks), and data variability (harmonization).
• Generative diffusion models enable cross-modal synthesis and prediction of missing data.

Conclusions:

• Despite significant advances, barriers like heterogeneous datasets, limited benchmarking, and regulatory hurdles impede clinical translation.
• Priorities include developing reliable, generalizable, and interpretable AI methods.
• These advancements are essential for progressing neuroimaging research and improving real-world patient care.