List of questions

What kind of new clinical imaging aspects will improve multi-disciplinary team meetings in the future?

Efficiency gain without giving up on quality is key during MDTs.
AI will suggest a diagnosis and annotate relevant images to present, CDS will recommend treatment based on similar patient data, but might also do this based on other imaging data/diagnosis and liquid biopsies.

Just like radiologists use imaging hanging protocols today for diagnosis, are there any substantial benefits creating something similar (e.g. AI driven image fusion and data composition) for collaborative care coordination of specific patient pathways?

Thanks to AI radiologists, pathologists, specialists will become collaborative clinicians working from anywhere. For patient centric personalized pathways, these clinicians will review relevant imaging and clinical data in team discussions. Is there an opportunity to fix these imaging and data elements in predefined layouts and formats?

What are the biggest challenges for AI in medical imaging to overcome towards integration into the current applications and workflows and how do you expect these to be tackled?

Most of the AI tools for imaging generate metadata stored into DICOM presentation state files which can be read by any PACS viewer. Is this sufficient? For every speciality a specifc AI tool might provide the best result. How do you see these being integrated into your viewer(s) and workflows?

Will the General Data Protection Regulation (GDPR) hinder industry: academic collaboration? Is academia ready for its implementation, particularly when running human clinical trials?

None

Quantifying tumour heterogeneity using texture analysis in CT and MRI has been demonstrated in many cancers. How feasible is texture analysis outside of the brain in PET & SPECT? In particular, prostate carcinoma diagnosis & staging using PET.

None

From academia’s perspective, what are the biggest barriers to working with industry?

None

1702: Radiomics, Radio-genomics, Pathomics. What are they, how do you validate these tools in a clinical trial and is there a future for these tools in a clinical (as opposed research or single–site) setting?

1739: How can deep-learning techniques be used to improve central radiological review in clinical trials within the next 5 years?

1751: What are the key pointers for a smooth central ARSAC approval process?

1702: None

1739:

1751: Often ARSAC applications can cause a substantial delay in the start up timelines of a trial. It is therefor in our interest to work with the ARSAC review team to reduce this as much as possible by being proactive and look at the risks prior to application.

What data set and expertise is necessary to implement a deep learning model on 3D medical images that can perform better than non-deep image processing and classifier methods?

At Brainomix, we develop neuroimaging software for diagnostic support using artificial intelligence approaches. Deep learning research his very popular but not always necessarily better than a simple k-means clustering algorithm. The reasons and the conditions for deep learning approaches to improve the performance of more classic machine learning techniques are often overlooked. Of a particular interest to us: What are the conditions for the design and training of a powerful deep learning 3D brain imaging diagnostic support tool?

Will clinicians trust machine learning approaches to assist decision making, especially if they are based on a ‘black box’ approach?

GE Healthcare

How big is big data, i.e. what is a meaningful dataset for either generating a hypothesis or validating it?

GE Healthcare

Where is AI on the Gartner hype cycle ( https://www.gartner.com/technology/research/methodologies/hype-cycle.jsp, https://www.gartner.com/smarterwithgartner/top-trends-in-the-gartner-hype-cycle-for-emerging-technologies-2017/)? What expectations can be met and which ones are unrealistic?

GE

Would a CD8 or CD3 PET tracer have clinical utility? How would it be used? With which drugs?

none

What are the barriers to clinical adoption of radiomics and how do we overcome those

none

Can deep-learning techniques benefit niche clinical imaging modalities where individual datasets are inherently small and methodology evolves over time and across sites?

How do we separate fibrosis from inflammation in the liver using clinical imaging?

How do we best assess tumour response to immunotherapy agents in clinical trials?

1702: Radiomics, Radio-genomics, Pathomics. What are they, how do you validate these tools in a clinical trial and is there a future for these tools in a clinical (as opposed research or single–site) setting?

1739: How can deep-learning techniques be used to improve central radiological review in clinical trials within the next 5 years?

1751: What are the key pointers for a smooth central ARSAC approval process?

1702: None

1739:

1751: Often ARSAC applications can cause a substantial delay in the start up timelines of a trial. It is therefor in our interest to work with the ARSAC review team to reduce this as much as possible by being proactive and look at the risks prior to application.

1702: Radiomics, Radio-genomics, Pathomics. What are they, how do you validate these tools in a clinical trial and is there a future for these tools in a clinical (as opposed research or single–site) setting?

1739: How can deep-learning techniques be used to improve central radiological review in clinical trials within the next 5 years?

1751: What are the key pointers for a smooth central ARSAC approval process?

1702: None

1739:

1751: Often ARSAC applications can cause a substantial delay in the start up timelines of a trial. It is therefor in our interest to work with the ARSAC review team to reduce this as much as possible by being proactive and look at the risks prior to application.

Is there a role for high throughput imaging in personalised medicine?

High throughput imaging has the potential to screen putative agents at various doses to find the optimum treatment - is it worthwhile?

How do we explore interactions of agents to find the optimum therapeutic window?

Synergistic interactions of drugs may offer increased efficacy, better tolerance and decreased adverse effects. However, the combinatorial effect blows the problem up to an unsolvable scale. How do we move from reductionist 1 dimensional assessments of utility up to poly-agent assessments?

How can academic collaboration help bridge the gap between technology validation and clinical adoption?

None

What are the biggest barriers to widespread adoption of digital pathology for routine clinical practice?

None

We are interested to learn of developments in HIFU specifically to image the stomach and lower GI. Is the presence of air and gas still a problem? If the patient drinks a litre of fluid just before the exam, is that sufficient?

Advances in Ultrasound Imaging of Human GI

How will AI related technologies in imaging have a direct impact on the NHS i.e. Patient Centric Outcomes and cost reduction

Understanding Clinical/Academic focus areas in Artificial Intelligence

In diagnostics what are the key clinical focus areas where imaging will make an impact

Understanding Clinical/Academic focus areas in Artificial Intelligence

What AI approaches are likely to have the biggest impact in clinical imaging

Understanding Clinical/Academic focus areas in Artificial Intelligence