What are the most effective strategies for optimizing the identification, analysis, reporting, and automatic follow-up of incidental pulmonary nodules to improve patient outcomes and streamline clinical management? (e.g. AI-based risk stratification)

Incidental pulmonary nodules are frequently encountered during imaging, posing challenges in terms of accurate identification, analysis, reporting, and subsequent follow-up. Optimizing the processes involved in the identification, analysis, reporting, and automatic follow-up of these nodules is crucial for improving patient outcomes and streamlining clinical management.

What strategies can be implemented to enhance efficiencies within lung cancer screening programs, with the goal of reducing the demand for healthcare human resources and capacity for low dose CT scans? (e.g. AI-based risk stratification, alternatives imaging modalities to low-dose CT scanning, risk stratification using Breathomics or liquid biopsy testing, etc.)

Lung cancer screening programs are essential for early detection and improved outcomes; however, the demand for healthcare human resources and low-dose CT scan capacity can present significant delays, ultimately impacting patient outcomes.

What would be the optimal targeted contrast agent for molecular imaging with Photoacoustics?

Photoacoustics (PA) imaging has been on the market as a preclinical technology for almost 15 years, with clinical applications and companies now emerging. There are many research groups and commercial organizations working on contrast agent development during this period, however, there still is not a good commercial solution to do true, reproducible, molecular imaging with PA. Many research papers exist that are application specific, but these seem to be limited to the groups that are experts in contrast agent development. Our question is really aimed at trying to find a truly universal and turn key contrast agent that could be used in preclinical molecular imaging studies of small animal disease models, to target and image cells of interest. This would be to try to get closer to what researchers are used to with optical imaging methodolgies like fluorenscence and biolumuninescence , but with the depth and resolution offered by photoacoustics.

How would you design a system to register ultrasound image data of small animals (e.g. mice) with other full body imaging modalities like MRI, CT, or optical?

Ultrasound imaging of small animals for study of human disease is now commonly used, but is inherently a 2D imaging modality with the option to scan in 3D by translating the imaging transducer laterally with a stage to build up a 3D data set. Ultrasound also requires gel or water coupling for imaging. Other modalities tend to be inherently 3D or tomographic (able to image the whole body of the mouse). There is a lot of interest in image fusion across multiple modalities. The question is aimed and strategies for being able to include ultrasound data in this fusion, either during image aquisition, or with post processing.

What are feasible scenarios and technical challenges for implementing MRI remote scanning?

Due to shortage of qualified MRTs, remote scanning may be an attractive option in certain circumstances, e.g. remote communities. We would like to discuss possible scenarios for implementing remote scanning as well as tackle possible challenges.

How can synthetic data be most efficiently utilized in training foundational healthcare AI models?

Access to sufficient amounts of clinical data to train foundational models may be challenging in many cases. Synthetic data produced by generative AI algorithms may be helpful to alleviate this problem but may also limit the model accuracy. We would like to discuss scenarios for using synthetic data in foundational model training.

Staff and equipment backlogs are a growing issue for many facilities. In the ICU, specifically, patient transport and radiation dose present unique challenges when/if other imaging is needed. such as CTs. What imaging strategies could help address these issues to improve both patient outcomes and hospital costs/efficiency?

We all know that our most critically ill patients are in the ICU – so if they need imaging, we must find a way to provide – even if it means moving the patient to another locale. Intrahospital transports come with high risks and costs and often over-extends the schedules on those other imaging devices. We would suggest that the optimal solution would be to bring better imaging capabilities to the bedside. Single exposure dual-energy subtraction on mobile x-ray devices has emerged as a strategy for enhancing bedside imaging in the ICU, with the potential to reduce the need of further imaging. Let’s expand this discussion and explore practical, easy-to-implement solutions that can deliver actionable results, as well as a better patient, staff and clinician experience in these (and other) critical care environments.

What is a good model in Canada and the US for remote screening which would reach underserved populations, using eye images to diagnose neurodegenerative diseases?

LumeNeuro has a patented novel, dye-free retinal imaging method that identifies protein deposits found in the retina as biomarkers of these deposits and of multiple neurodegenerative diseases in the brain, including Alzheimer’s disease. This test can detect and identify deposits of different protein types as differentiators of brain diseases, years before first symptoms. The technology won 2nd place in an Alzheimer’s Association pitch competition because of its potential to reach underserved populations, through the remote assessment of images.
In Ontario, the expansion of telehealth services has been slower than was anticipated. In both Canada and the US, eye screening in remote communities is still often delivered using travelling teams. In Canadian and US cities, community centres are used for the intensive screening of underserved populations. But this coverage is sparse. What model would improve such coverage?

How to enable robust and reliable magnetic resonance imaging of the brain without sedation in a pediatric population aged 2–4-year-old, in a clinical research context

Magnetic resonance imaging (MRI) plays a central role in the diagnostic and therapeutic pathways for multiple pediatric diseases, avoiding the ionizing radiation concern associated with some other imaging modalities. However, MRI is comparatively slow and motion-sensitive, which poses challenges when imaging pediatric populations. While a “Feed and Swaddle” technique can effectively be used when imaging neonates and young babies, children in the 1- to 5-year-old age range are particularly difficult to image without resorting to sedation.

Can an AI-based Radiologist Assistant app be developed to easily present clinical data to assist radiologists when interpreting imaging exams, provide differential diagnoses based on imaging findings described in the report, and automatically calculate RADS categories (malignancy risk categories for various lesions based on set criteria)?

n/a

When measuring blood pressure from left and right arms, the readings can change significantly for some subjects, what are the factors that affect the lateral difference in measurements. How does the lateral difference change during the day, week, month and longer periods. Do external factors such as exercise, medication, temperature differences etc. change the lateral difference?

Hypertension is a highly prevalent (~45% adults in the US) risk factor for cardiovascular diseases. Frequent measurements of blood pressure (BP) in an ambulatory setting are necessary to assess the cardiovascular mortality risk. BP measurements in the clinics can sometimes lead to higher numbers due to the ”white coat syndrome” attributed to the anxiety associated with the clinical setting. The current standard for at-home BP measurements is a cuff-based device that can be quite discomfortable for subjects with high BP as well as during the night. This can lead to reduced compliance in BP measurements. Moreover, BP measurements can vary significantly laterally between different arms. It is important to understand what factors affect this lateral difference and how it changes over time or due to other external factors for cuffless BP measurement devices.

Why does the photoplethysmography (PPG) signal appear inverted when measuring at certain anatomical sites?

PPG signal origin has been attributed to the changes in attenuation of light during the cardiac cycle at the measurement site. At some anatomical sites, measured PPG signal appears to be inverted, what is the reason for the inversion of PPG waveform?

Q: Is the promise of GenAI in any of these contexts realistic? These generative models tend to create low resolution features. Can we extend these foundational models to improve their utility, eg encoding higher resolution? Specifically, for realistic phase-retrieval/synthesis from magnitude DICOMs, is GenAI a possible solution?

Part A: When training other deep learning image enhancement networks (denoising, super resolution, artifact reduction, motion compensation, etc), we often run into the problem of having limited data sets (limited in anatomy/physiology, pathology, populations, etc..). Generative AI is an option for synthesis of training data, as has been demonstrated by the NVIDIA MONAI collaboration with Kings College London (Project-MONAI/GenerativeModels: MONAI Generative Models makes it easy to train, evaluate, and deploy generative models and related applications ). This project could look at implementing and fine tuning a generative model, based on RadImageNet, that will serve to expand our datasets for training, thereby reducing risk, cost, and time to market for our deep learning imaging solutions

Part B: MRI data is complex-valued, but typically output images are shown and saved in magnitude and the phase information is discarded. Often having the complex-valued data allows us to do significantly better image processing. The phase is well understood and a function of the magnetic susceptibility of the tissues, and the tissue interfaces and their orientation with respect to the main magnetic field.

Q: Can deep learning techniques be applied to fat suppression? Both in terms of the complete solution, especially in the vicinity of large background magnetic field inhomogeneity and/or as an accelerator for CSE processing.

Specific clinical application: Imaging of the orbits when the subject has braces.

There are many MRI applications that require the suppression of signal from fat. Typically, this is done with a separate saturation pulse (“fat-sat”) that is used to reduce the signal from fat prior to acquisition, or by selective excitation (exciting only the water signal). These methods of fat suppression are much harder at 0.5 T since the distance between fat and water peaks is only 3 ppm (70 Hz at 0.5 T). Another fat suppression method that we have been implementing at Synaptive is by chemical shift encoding (CSE) the image with multiple echoes; however, this method requires a lot of computation and a relatively slow acquisition. To make matters worse, all fat suppression methods tend to fail when in the vicinity of large magnetic field inhomogeneities. This is less of a problem at mid-field, except when metallic devices are present (e.g. braces).

Q: What post-processing solutions are available for motion correction? Are they effective? Will they work at 0.5 T?

Motion corruption of images is a common problem in MRI, especially when imaging patient populations with neurological disorders. The artifacts associated with motion corruption can change depending on the pulse sequence, but most commonly appear as “ghosting” and blurring of the image.

Cardiac volume, flow and pressure are related. Given the volume and change in volume of a chamber over the cardiac cycle, it is possible to model the in and out flow to the same accuracy and definition as the gold-standard (2D echo doppler)? What are the limitations? What are the assumptions?

When assessing the function of the heart, consultant cardiologists look to quantify volume, flow and pressure in the four chambers of the heart. However, there is no single ideal technology that can accurately, reliably and cost-effectively do this. Currently, cardiologists have to use several medical devices including 2D echo, 3D echo, cardiac MRI, CT and pressure catheterisation.
As these technologies use different ground-zeroes, combining these datasets to provide an instantaneous whole-heart functional measure is not possible at the clinical level.
Ventripoint’s VMS+ accurately measures the volume of all four cardiac chambers from 2D echo images (vs gold-standard volume measurement using cMRI) across the complete cardiac cycle.
VMS+ uses Knowledge-Based Reconstruction to recreate heart chambers leveraging an extensive database of cMRI images.

Cardiac volume, flow and pressure are related. Given the volume and change in volume of a chamber over the cardiac cycle, it is possible to model the pressure to the same accuracy and definition as the gold-standard (pressure catheter)? What are the limitations? What are the assumptions?

When assessing the function of the heart, consultant cardiologists look to quantify volume, flow and pressure in the four chambers of the heart. However, there is no single ideal technology that can accurately, reliably and cost-effectively do this. Currently, cardiologists have to use several medical devices including 2D echo, 3D echo, cardiac MRI, CT and pressure catheterisation.
As these technologies use different ground-zeroes, combining these datasets to provide an instantaneous whole-heart functional measure is not possible at the clinical level.
Ventripoint’s VMS+ accurately measures the volume of all four cardiac chambers from 2D echo images (vs gold-standard volume measurement using cMRI) across the complete cardiac cycle.
VMS+ uses Knowledge-Based Reconstruction to recreate heart chambers leveraging an extensive database of cMRI images.