AI-Enhanced Digital Health: How AI Can Make a Difference for Personalized Therapy

The shift towards digitalization in healthcare and the advancement of AI technologies have opened up opportunities for digital health companies, as well as for pharmaceutical and medical device corporations seeking to improve their product offerings. Businesses aiming to leverage new technologies are searching for effective ways to integrate AI into their products while ensuring full compliance with regulations.

DataArt recently hosted a webinar to explore successful examples of AI technology integration, best practices, and the regulatory hurdles associated with AI-enhanced digital health.

The panel of experts moderated by Ivan Pantykin, VP of Healthcare and Life Sciences, DataArt, included:

• Patrik Scholler, Strategy Consultant, Life Sciences Consulting
• Andrey Sorokin, AI Expert and Solution Architect for Healthcare and Life Science Practice, DataArt
• Olver Hilgers, Head of Digital Health, Regenhold

In this article, we summarized the key insights from their discussion.

AI-Enhanced Healthcare Products: Next-Generation Solutions

AI can potentially empower various areas of healthcare, one of them being personalized cancer therapy. AI-based software can identify specific cancer subtypes and genetic alterations and enable doctors to select targeted therapies with better efficiency and fewer side effects.

AI technologies can also be used to build an app that continuously measures a brain's electroencephalogram (EEG) with an ear sensor to predict seizures in epilepsy patients. The app could set alarms before an upcoming seizure and give patients time to find a safe space and prepare. Additionally, the app could collect various data, such as stress levels and dietary information, to support AI-based learning and enhance therapy management.

Another innovative concept involves integrating an intelligent chatbot into DiGA (reimbursable health apps) to provide disease and therapy-related support to patients. This chatbot could offer medication reminders, track symptoms via voice input, and provide regular health coaching.

"Patients recognize the immediate value for their therapy journey while using these apps, so they are enthusiastic about it and willing to provide their data. This can change the way people approach therapy and handle their personal information. Of course, the information must be well-protected, as building trust is crucial. But once it’s done, this data can be incredibly valuable, not only for individual patients and their therapy but also for the broader medical community and scientific research." - Patrik Scholler, Strategy Consultant, Life Sciences Consulting.

Prerequisites and Best Practices for Developing AI Modules in Medical Diagnostics and Device Software

AI technology serves as a powerful augmentation of human capabilities in the field of healthcare, offering unprecedented reaction times and scalability. The scalability of AI solutions is attainable through cloud technologies, local data center deployment, or edge devices, depending on the application.

The use of larger language models (LLMs) has spread in many industries, with models like ChatGPT demonstrating the power of generative pre-trained transformers. Prompt engineering and retrieval-augmented generation (RAG) enable real-time adaptation of model behavior and reduce the risk of generating irrelevant or inaccurate information.

In addition to LLMs, there is a growing trend towards small language models (SLMs) tailored with specific domain knowledge. SLMs require significantly less data than LLMs and can be run locally, even on smartphones. This approach enhances data privacy by keeping sensitive information within local environments, reducing the potential surface of attack.

"Small language models have the potential to become a game changer in data privacy as they can be exposed to sensitive data that will not be transferred anywhere outside the hospital." - Andrey Sorokin, AI Expert & Solution Architect, DataArt.

Advancements in computer vision and image analysis also significantly impacted healthcare. Multimodal machine learning solutions can translate medical images directly into textual descriptions, providing a new level of insight and understanding. These solutions can also utilize multimodal data, such as medical records and video recordings, to detect early signs of conditions like neurodegenerative diseases. Furthermore, these advanced models can be deployed on smart devices, reducing reliance on cloud technology. This is particularly crucial for medical devices, ensuring reliability and functionality even in environments with limited or no network connectivity, such as during surgery or emergency situations.

Navigating Regulatory Challenges in AI-Enhanced Medical Devices

Businesses seeking to integrate AI into their medical devices need to consider how to properly address regulatory requirements and structure the process for certification. These requirements were partly addressed in the recent European Union's AI Act.

"The EU AI Act doesn’t establish any particularly new requirements. You just need to do proper technical AI technology and risk assessment." - Olver Hilgers, Head of Digital Health, Regenold.

Contrary to common misconceptions, the use of AI in medical devices does not automatically influence their classification. The AI Act applies to Class II devices, while for Class I devices, it is recommended to follow AI Act requirements as "state of the art" without certification.

Using Large Language Models (LLMs) in medical diagnostics presents certain challenges, particularly in reproducing exact content. Measures against fabricating, such as human oversight, explainability, Retrieval Augmented Generation (RAG), transparency, and semantic entropy, are crucial for ensuring the reliability and accuracy of AI-driven outputs.

Clarity in regulatory definitions and certification requirements is essential for medical device manufacturers utilizing AI. It is important to understand the impact of AI on product classification and certification under the AI Act. Engaging with different notified bodies and seeking their expertise can provide valuable insights and support in navigating the regulatory landscape.

Collaborative Approach in AI-Enhanced Healthcare: Navigating Complexity for Success

No single entity possesses all the necessary competencies to build an AI-enhanced digital health product from the beginning to the end. Clearly, a collaborative approach is necessary to successfully integrate AI into healthcare solutions due to the complexity and diverse expertise required. That’s why DataArt consolidated its expertise with a number of highly competent parties with a proven track record to build the AIeHC consortium.

The primary purpose of the consortium is to help and empower clients in realizing their healthcare vision by providing guidance and support. This involves conducting a thorough reality check and business analysis, overseeing product design and development, and navigating technical and clinical validation, including regulatory approval, to prepare for market introduction.

Conclusion

Businesses can leverage AI to enhance the precision, efficiency, and safety of digital health products if they embrace best practices for AI module development and successfully navigate regulatory challenges while using a collaborative approach.

You can watch the full webinar and get more insights on AI-enhanced digital products here:

If you have an idea for a digital health product, contact us to receive professional guidance and support from our team of experts.

Originally published here .