What is IXIQ.Ai? Find out from our Chief Scientific Officer Robin Wolz

This month, we launch into production on its first clinical trials. IXIQ.Ai is the next generation of our technology platform which we have been developing for more than a decade.

With novel deep-learning AI tools and the increasing numbers of curated datasets at our disposal, we can now automate the accurate measurement of some of the most complex brain structures that play a key role in understanding neurodegenerative diseases such as Alzheimer’s disease and Huntington’s disease.

Our Chief Scientific Officer, Dr Robin Wolz, has led the development of IXIQ.Ai with our science team for the last two years. We spoke to him about the evolution of the new platform, the process his team have gone through, and the exciting potential for clinical and healthcare applications both now and in the future.

Robin, what exactly is IXIQ.Ai?

First, a lot of people don’t realise IXICO stands for Information eXtraction from Imaging. And the ‘Q’ in IXIQ.Ai stands for Quantitative.

In summary, IXIQ.Ai is an AI-based platform for brain segmentation. It gives our development engineers an agile infrastructure which enables us to efficiently and rapidly deploy imaging analysis solutions.

IXIQ.Ai can be trained on a specific problem by feeding the platform with training data for a specific brain region in a specific therapeutic indication. We call the resulting algorithm a ‘plugin’ that can be deployed on our data management system, TrialTracker, to perform analysis on clinical trial data.

Taking Alzheimer’s Disease (AD) and Huntington’s Disease (HD) as examples, there are different brain regions relevant in the two indications, and we need to train the platform on data from the specific population. Taking young, healthy subjects wouldn’t work as well as it wouldn’t be able to learn the patterns of neurodegeneration observed in the different patient populations.

Currently, the platform is pre-trained on our extensive database of clinical trial and natural history datasets across several CNS indications. Through the pre-trained models in the IXIQ.Ai platform infrastructure, plugins can be trained with a reduced number of highly curated datasets to obtain an optimised solution for a specific segmentation problem.

The validation we have performed over the past years in developing plugins for HD and AD has demonstrated IXIQ.Ai offers lower QC failure rates and reduced volume error relative to manual raters when compared to widely used tools. This means more data can be analysed in a clinical trial with increased accuracy.

How long has IXIQ.Ai been in development?

We have been working on this specific project for around two years with a dedicated team of engineers – though it builds on the expertise, know-how and data we have collected at IXICO over more than a decade.

What did the development process for IXIQ.Ai look like?

As we are so closely embedded within the ecosystem of Central Nervous System (CNS) clinical research, we have been all too aware of the shortcomings of more traditional analysis techniques – especially when it comes to the segmentation of more complex brain structures.

Particularly in HD, the striatal regions of the brain are very difficult to segment because there is a very low contrast between those regions and the surrounding brain tissue. Even for a manual rater it’s not straightforward, and this is where traditional techniques fall short.

Our development process for IXIQ.Ai had three key stages:

1. Development of the hardware and software infrastructure to perform image segmentation based on convolutional neural networks (CNNs) at scale. In this phase, our technology team set up the server infrastructure, specifically the Graphic Processing Units (GPUs) which are the computer processors used to solve neural networks. We then built the software infrastructure (the neural network architecture) that forms the core of the IXIQ.Ai platform and that can take data in to train specific plugins.   
2. Curation of datasets in our core indications to train the platform on specific instances. Over the past 5-10 years we have built up a database of clinical trial and natural history datasets across the core indications we work in. Leveraging this data to train IXIQ.Ai plugins requires curation and annotation of the data. As part of the development process, we have now completed this for our AD and HD datasets.
3. Validation of the developed instances with our academic and pharma partners. Once development of the AD and HD plugins was completed, significant effort went into validating the new technology with our academic and pharma partners to show how IXIQ.Ai increases performance over currently available tools. We have jointly presented those results at numerous conferences with our collaborators.

We are looking forward to presenting further findings with our academic and pharma partners at the CHDI 17^th Annual HD Therapeutics Conference in Palm Springs, California, as well as the 15^th ADPD Conference in Barcelona, Spain.

How is IXIQ.Ai an evolution from LEAP?

 IXIQ.Ai enables us to measure things we couldn’t measure before. It sets a new standard compared to LEAP and other widely used platforms such as FreeSurfer. We set new standards with LEAP 10 years ago, and now we’re raising that standard again.

 The main advantages of IXIQ.Ai over traditional machine learning techniques are:

1. CNN-based approaches offer an increased ability to ‘learn’ more subtle or challenging data patterns. In practical terms, this makes the platform applicable to more difficult-to-measure brain structures like the putamen in HD.
2. A significant reduction in compute time to analyse one scan (seconds vs hours). Traditional tools like FreeSurfer and LEAP run for several hours to segment the brain. IXIQ.Ai takes seconds to minutes to perform the same task, which is an incredible difference. This significantly shortens our development cycles, as the time reduction in analyzing hundreds of data sets is huge, as well as allowing for more efficient deployment in production. This opens up a wealth of potential opportunities for clinical application where real-time processing is a requirement.

How is IXIQ.Ai different to what's out there already?

The concept of a neural network-based approach is not new or exclusive to IXIQ.Ai. Beyond the optimisation of such an approach to 3D brain segmentation, a key value lies in the combination of this approach with the data we have collected which is essential to train the platform.

You might have heard people say, “Data is the new oil” or “Data is the new source code” and that’s absolutely right. When you have the data, you can develop a proprietary algorithm that becomes your ‘secret sauce’.

Our team has spent the last decade collecting and curating a vast number of unique datasets from clinical trial and natural history studies which we have the rights to use as part of our R&D efforts. It is this highly curated and contextualised data which allows us to unlock the full potential of the underlying AI architecture.

What are the clinical applications of IXIQ.Ai in your key therapeutic areas?

IXIQ.Ai provides volumetric segmentation in different brain regions affected in different therapeutic indications, such as the hippocampus and ventricles in AD and the putamen and caudate in HD.

The fact that we can measure volumes in those brain regions enables us to determine patient eligibility criteria for clinical trials. For example, in studies where the drug is administered directly to the brain, we need to understand the size of the brain structure before drug administration.

The platform can also map changes to brain volumes over time, allowing us to measure the long-term efficacy of treatments more accurately. For example, with AD, the hippocampus – which is responsible for short term memory – shrinks, therefore you would expect an efficacious drug to slow that progression and result in a decline in shrinkage. Such changes can be tracked with IXIQ.Ai.

What is the key impact you hope IXIQ.Ai will make in clinical research, and for patients?

With IXIQ.Ai we have a higher chance of identifying a treatment effect in a clinical trial. As we can analyse more data, we gain more statistical power and a more accurate measurement. The platform is sensitive to small changes, and we can look at brain regions that we couldn’t previously because we couldn’t measure them accurately in an automated way. We also expect it will give us additional biomarkers to analyze.

We hope an improved imaging measure will contribute to expediting the drug development process and therefore play an important role in making new and effective treatments available sooner for patients with neurodegenerative diseases.

Want more information on IXIQ.Ai?

To request further information or to arrange a demonstration, please contact us

---