Autonomous Magnetic Resonance Imaging, or AMRI, is meant to operate end-to-end. What this means is that, AMRI should intelligently acquire, reconstruct and visualize MR data.
We also introduce AMRI-Scanner as an Online Service (AMRI-SOS). AMRI-SOS is a free to use online service. Check out the amri-sos
branch to learn more, or fill out this form to request AMRI-SOS scans.
Globally, there are 4.6 MRI scanners per 1,000,000 people. In comparison, there are 36.72 per 1,000,000 in the United States [1][2], and the low-income countries have 0.1 per 1,000,000. Annual Healthcare spending in the US is 700 billion USD. Of this 125 – 200 billion USD (at least) is accounted for by diagnostic imaging [3]. Autonomous scanners can play a major role in tackling the challenges of delivering MR to a larger population with low-incomes. [4].
MR value can be defined to the patient as the ratio of actionable diagnostic information to the costs or time incurred. A simplified and technically relevant definition of MR value is: ratio of contrast-to-noise ratio (CNR) to the total scanner (table) time.
AMRI is capable of transforming a conventional MR system into an IPS that can be characterised by the following capabilities:
- Cognizant
- Taskable
- Adaptive
- Ethical
- Reflective (Reuse, Retain, Revise)
The above capabilities would deliver an IPS that is knowledge-rich, and capable of long-term autonomy with minimal human intervention and supervision.
AMRI can be easily built, rebuilt and deployed as it leverages 4 existing file standards and introduces 1 new file standard. These file standards are:
- EMR file standard for subject information encryption
- Situation report (Sitrep) file standard for sampling MR system state (introduced in this work)
- Pulseq [5, 6, 7] file standard for multi-vendor pulse sequence programming
- ISMRMRD file standard for acquired raw data
- DICOM file standard for reconstructed image data
AMRI can also enable self administered MR scanning with the inclusion of MR-safe A/V accessories.
YouTube demos can be found here:
- Invivo experiment, link
- Phantom experiment, [link][invivo-exp-2]
google-cloud-speech google-cloud-texttospeech keras numpy Pillow pyaudio pyautogui pygame pandas pydrive pypulseq scipy tensorflow tqdm
👉 AMRI-SOS form 👈
AMRI-SOS enables students and researchers with limited access to MRI hardware to leverage AMRI to run custom pulse sequences designed on the Pulseq [5, 6, 7] framework. If you want to run your scans on a Siemens Prisma 3T, fill out this form.
[1] Atlas of MS 2013, https://www.msif.org/wp-content/uploads/2014/09/Atlas-of-MS.pdf
[2] Geethanath S, Vaughan Jr JT. Accessible magnetic resonance imaging: A review. Journal of Magnetic Resonance Imaging. 2019 Jan 14.
[3] Kelley, Robert. "Where can $700 billion in waste be cut annually from the US healthcare system." Ann Arbor, MI: Thomson Reuters 24 (2009).
[4] Wang, Ge, et al. "Image Reconstruction is a New Frontier of Machine Learning." IEEE transactions on medical imaging 37.6 (2018): 1289-1296.
[6] Ravi, Keerthi, Sairam Geethanath, and John Vaughan. "PyPulseq: A Python Package for MRI Pulse Sequence Design." Journal of Open Source Software 4.42 (2019): 1725.