A foundational AI-MRI framework for self-supervised discovery of transformative low-field MRI techniques

Descripción del proyecto

Resonancia magnética mejorada con inteligencia artificial

La resonancia magnética, técnica no invasiva que utiliza campos magnéticos y ondas de radio, produce imágenes de alta resolución de tejidos y órganos. Puede utilizarse para diagnosticar enfermedades sin exposición a la radiación. Sin embargo, su elevado coste limita su uso clínico. La resonancia magnética de bajo campo funciona con campos magnéticos de menor intensidad y promete imágenes asequibles y portátiles, pero se resiente de la calidad de imagen. El proyecto NEW-CONTRAST-MRI, financiado por el Consejo Europeo de Investigación, propone integrar la inteligencia artificial en un marco autosupervisado para optimizar los parámetros de ensayo. Se espera que esto acelere la exploración y mejore la calidad y el contraste de las imágenes, transformando así la resonancia magnética de bajo campo en una modalidad adecuada para entornos clínicos con pocos recursos.

Objetivo

Magnetic resonance imaging (MRI) is crucial to healthcare for its radiation-free, high-quality scans, but its high cost leaves 70% of the global population without access. Emerging low-field MRI technology offers affordable, portable systems with transformative potential, but faces critical challenges: long scan times, low signal-to-noise ratio (SNR), and poor tissue contrast, which makes some tissues indistinguishable. Although contrast agents help, they add risks. Early studies showed that low-field MRI can detect cancer without contrast agents using unique pulse sequences, but those are manually designed and slow. Recently, AI has been adopted for clinical high-field MRI, but AI pulse-sequence optimization relies on supervised learning, limiting discovery, AI theory is scarce, and simulations do not fully capture MRI’s complex spin dynamics. Moreover, AI use in low-field MRI remains mostly focused on image post-processing, while pulse sequence design, sampling, and reconstruction remain traditional and suboptimal.

I propose to develop a foundational AI framework that will transform low-field MRI into a rapid, high-quality modality. To break the barriers posed by supervised AI and simulations, my key innovation is an integrated AI-MRI framework, where on-the-fly MRI measurements guide a self-supervised AI search through the parameter space. Leveraging AI foundations I have recently developed, the framework will jointly optimize pulse sequences, sampling, and reconstruction to revolutionize imaging. Specific aims: (1) speed up MRI by an order of magnitude; (2) establish AI theory; (3) build the framework and develop cutting-edge sequences for optimal tissue contrast; and (4) demonstrate these in human scans with my lab’s low-field MRI. Preliminary results support the feasibility of our design aims. This project will transform low-field MRI into a fast, affordable, contrast-agent-free tool with broad clinical applications, particularly in low-income regions.