Published on September 21, 20213 min read
By Rabindra Khadka
Clearbox AI Control Room tackling the AI4EU Challenge

Clearbox AI strives to harness powerful AI technologies to improve businesses and society in a trustworthy way. Recently, we competed and won the pan European AI4EU call for challenges and successfully proposed our solution to enhance clinical AI work flows at the hospital CHU de Liege, Belgium using our AI Control Room platform.

What is AI4EU?

AI4EU is a European R&D initiative to build an on-demand platform to create and sustain an AI ecosystem in Europe with stakeholders not limited to SMEs, industries, academia, funding organizations and citizens. The AI4EU consortium published a call for challenges related to AI after collecting the business or operational changes faced by companies and organizations while adopting AI technologies. The challenge that we successfully proposed our solution was posed by the Belgian hospital on enhancing their clinical AI workflows.

Why are AI workflows a challenge?

CHU de Liege produces huge volumes of data at different departments of the hospital and wishes to extract valuable insights from this data using AI models to improve patient outcomes and experience. In the process of achieving these objectives, the hospital recognised two challenging paths to tread i) Developing a data integration process of the disjointed data across departments to the institutional data warehouse. ii) Development of Clinical Decision support system (CDSS) using artificial intelligence processes based on real-time data from various departments. The ultimate challenge is to apply artificial intelligence while making use of structured and unstructured data from various hospital departments and finally help clinicians to make prompt and correct decisions during practices.

Clearbox AI’s focus in the challenge

Clearbox AI’s contribution will focus on assessing, explaining and improving the CDSS. The solution also tries to envisage the notion of humans in the loop by allowing clinicians to be part of CDSS by leveraging the conclusion drawn by an explainable AI system. Our solution will not only improve patient outcomes but also increase the trustworthiness of the AI models from the clinician’s point of view.

Explainable and multi-modal CDSS powered by Clearbox AI Control Room

The solution proposed by Clearbox AI takes into account the multimodal data available from different departments at the hospital. We propose a multi-modal AI system that can co-align extracted features and train a deep neural network for classifying normal or pathological cases. Furthermore, given Clearbox AI’s expertise in the field of explainable AI, the proposed solution contributes to the clinical decision support system (CDSS) that uses the in-house explainable AI engine which is part of Clearbox AI Control Room. The solution captures the idea of providing interpretable results to clinicians and involves them in the decision-making process by quantifying the uncertainty associated with AI prediction.

A good example of MLOps in action for AI assessment and monitoring

Clearbox AI Control Room technology bridges the gap between the prediction and post-prediction stages. The core part of the technology focuses on dataset assessment, explainability, quantification of errors, monitoring, and setting up the platform for active learning. The AI Control Room seamlessly integrates into the CDSS of the hospital as it is model agnostic. It is also part of the data analysis tool that analyses and presents the data analysis in a visual format for the practitioners in the CDSS platform. We will continue to provide updates on the solution progress after testing and deployment.

Enhancing AI work flows is an excellent use case that highlights the importance of AI assessment and monitoring in organizations planning to leverage data driven decision making using AI. We are constantly working on honing our AI Control Room platform to assist data scientists irrespective of the sector they are working in, to facilitate their workflows and offer a convenient way of assessing, interpreting, monitoring and deploying their AI models.

Rabindra Khadka is a Data Scientist at Clearbox AI. In this blog, he writes about Machine Learning models training and validation.