Artificial Intelligence and PET/MRI

Artificial Intelligence and PET/MRI
Manjari Murthy, Director of Product Management

Artificial intelligence, or AI, is one of buzziest topics in medicine. AI in medicine is a blanket term to describe computer programs that use machine learning to improve doctors’ ability assess medical data with the goal of improving health outcomes. This can be in the form of data processing, image interpretation, computer-assisted diagnosis, or a host of other applications. AI has been particularly important in radiology, where advanced algorithms have contributed to, for example, the identification of subtle pathologies, leading to earlier diagnosis and potentially better treatment options.

In brain PET/MRI, AI has been applied in a number of innovative ways, from improving very technical aspects of the imaging process, to more forward-facing components of image presentation and analysis. Data from the PET and MRI scans work in concert during the image reconstruction process through help account for attenuation, or dampening, of the PET signal as it passes through materials of different density in the brain – tissue, bone – to correct distortions and create a more accurate image. Deep learning, a type of AI that can be ‘trained’ to precisely identify distinct features within a data set, has been used to improve attenuation correction by discriminating finer details within images acquired. This provides a more accurate, more detailed map of differences in brain density, thereby improving image quality (Metecher et al, 2020). Deep learning has also been applied by researchers to generate PET/MR images using low doses of radioactivity that are of equivalent quality as images obtained using higher doses (Chen et al, 2021). From an image interpretation perspective, AI has been applied in the analysis of PET/MR brain scans of older adult patients to separate those with no brain abnormalities, those with hallmarks of mild cognitive impairment, and those with Alzheimer’s Disease (Liu et al, 2018). This can be used in concert with clinical data to support a diagnosis and optimize treatment.

The benefits of the application of AI to PET/MRI scans to patients is tremendous, and would be a very valuable addition to the Cubresa BrainPET, one which we intend to explore in the future. Lower dosing requirements to get clinical quality data can be a significant cost saver for hospitals, as the radioactivity is a large part of the cost of a PET/MR scan. Lower doses also means that pediatric patients could be scanned more safely – lower radioactivity administration poses less of a risk – and that patients could be scanned more frequently, allowing PET/MRI to be used as an observational tool over a longer period of time. Improvements in the use of imaging data means that the volume of data required to generate a clinical-quality image would be less and that scans could be done more quicky. This would be a huge benefit to patients who may be distressed during a scan and could only tolerate short sessions. In a similar vein, improved ability to use data that might otherwise have been discarded due to poor quality means that patient movement could be compensated for. This would be a huge boon to patients with tremors or movement disorders, who could benefit from the power of PET/MRI; coupled with the added benefits of BrainPET to clinicians and scientists, such as lower cost and retrofit nature, it could prove to be a sea-change in brain imaging.

AI is an exciting new frontier in brain imaging using PET/MRI, and BrainPET in particular, and as collaborations between the fields of computer science, medicine, and engineering evolve, revolutions in brain imaging, and innovations and creativity in its use, are bound to follow.

Mecheter, I., Alic, L., Abbod, M. et al. MR Image-Based Attenuation Correction of Brain PET Imaging: Review of Literature on Machine Learning Approaches for Segmentation. J Digit Imaging 33, 1224–1241 (2020)
Chen, K.T., Toueg, T.N., Koran, M.E.I. et al. True ultra-low-dose amyloid PET/MRI enhanced with deep learning for clinical interpretation. Eur J Nucl Med Mol Imaging 48, 2416–2425 (2021)
Liu, X., et al. Use of multimodality imaging and artificial intelligence for diagnosis and prognosis of early stages of Alzheimer’s disease. Transl Res 194, 56-67 (2018)