🚨 Pulse of Innovation 🚨

Breaking news in the healthcare AI

TWO years of chatGPT and Generative AI revolution: Impact on Healthcare

Open AI’s Large Language Model (LLM) based Natural Language Processing (NLP) has given rise to generative artificial intelligence (AI) and that has powered the generative pre-trained transformer (GPT) models that are modified as chatbot to match the human conversational skills and that has become chatGPT. Since its beginning several upgrades have been made and the latest chatGPT 4o is considered the market leader in Generative AI. chatGPT has competitors such as Google’s Gemini, Meta’s LLaMa and others but is considered the market leader at this point.

One of the most promising areas of application is in patient communication and engagement. ChatGPT can be used to create personalized health information, answer patient questions, and provide support and encouragement. This can help to improve patient satisfaction and adherence to treatment plans. For example, ChatGPT can generate tailored educational materials based on a patient's specific condition, making complex medical information easier to understand. Additionally, it can be used to create interactive chatbots that can answer patient questions 24/7, reducing the burden on healthcare providers.

Another area where ChatGPT is showing promise is in clinical decision support. By analyzing vast amounts of medical literature and patient data, ChatGPT can provide clinicians with evidence-based recommendations. This can help to improve diagnostic accuracy and treatment decisions. For example, ChatGPT can identify potential drug interactions or recommend alternative treatments based on a patient's specific characteristics. Additionally, it can be used to generate clinical reports, saving clinicians time and reducing the risk of errors.

However, the use of ChatGPT in healthcare also raises concerns about data privacy, security, and the potential for bias in AI algorithms. It is crucial to ensure that patient data is protected and that AI systems are trained on diverse and representative datasets to avoid perpetuating harmful biases.

In conclusion, ChatGPT has the potential to revolutionize healthcare by improving patient engagement, clinical decision support, and administrative efficiency. However, it is essential to address the ethical and technical challenges associated with its use to ensure that it benefits patients and healthcare providers without compromising safety and equity.

These are still considered early days of generative AI and its real-world application is taking steps in unknown territory of mix of human and artificial intelligence. Human control is critical especially in the field of healthcare not only from patient privacy but also from the patient well-being aspect. For this reasons, a hybrid human health intelligence could be considered the gatekeeper as going forward to provide safest and cutting technology driven generative Health AI.

🧑🏼‍🔬 Bench to Bedside👨🏽‍🔬

Developments in healthcare AI research and innovations

Advancing Heart Failure Management with AI-Driven NLP

Background 

• Heart failure (HF) management depends on functional status assessments like the New York Heart Association (NYHA) classification, which guides treatment and trial eligibility. However, these assessments are often unstructured in clinical notes, limiting their utility.

• A recent study by Adejumo et al. proposes an AI-driven natural language processing (NLP) approach to extract and enhance the use of this critical data.

Key Features of the Study

• The research analyzed EHR data from 34,070 HF patients across three healthcare networks in Connecticut.

• Technology: Fine-tuned ClinicalBERT to detect explicit NYHA classes and categorize functional status from HF symptom descriptions.

• Validation: The model was tested across diverse settings, showing excellent performance with AUROCs of 0.98–0.99 for NYHA classification and 0.94–0.95 for symptom categorization.

Results 

• The NLP model demonstrated robust performance, with AUROCs of 0.98–0.99 for detecting NYHA classes and 0.94–0.95 for identifying HF symptoms during activity or rest.

• Among 182,308 unannotated notes, the model identified explicit NYHA mentions in 13.1% of cases. An additional 10.8% were categorized into NYHA classes based on symptom descriptions.

• Overall, functional status classification increased by 83% when combining explicit mentions with model-derived classifications.

Clinical Impact 

• Improved Documentation: Enhanced the capture of HF severity to support guideline-directed care.

• Decision Support: Identified patients needing urgent interventions, such as advanced HF therapies.

• Research Advancement: Streamlined trial recruitment by accurately screening large patient cohorts.

• Scalability: Demonstrated reliable performance across diverse clinical settings.

Limitations 

• The model’s performance may vary with institution-specific documentation practices, and differentiating NYHA classes II and III remains a challenge due to overlapping clinical descriptions.

Conclusion 

This study highlights the potential of AI-driven NLP to bridge gaps in HF management by transforming unstructured data into actionable insights, enabling better care delivery and research efficiency.

JAMA Article

Radiology AI platform Deepc.ai : Beyond AI powered diagnostics

Deepc.ai is a company focused on revolutionizing radiology workflows through artificial intelligence (AI). Their core product, deepcOS®, is an "AI Operating System" designed to integrate seamlessly with existing healthcare IT systems. This allows radiologists to leverage the power of AI directly within their current workflow, eliminating the need for major infrastructure changes.

Here's how deepc.ai aims to improve radiology:

• Efficiency: deepcOS® integrates with any radiology AI solution, streamlining reporting processes. It can automatically populate radiology reports with AI-driven findings, saving radiologists valuable time.

• Accuracy: By utilizing AI for tasks like identifying abnormalities, deepcOS® can potentially improve the accuracy of radiology reports.

• Focus on Patient Care: By reducing time spent on administrative tasks, deepcOS® frees up radiologists to focus on patient care and complex cases.

• Scalability: deepcOS® is designed to be flexible and adaptable, working with a wide range of AI solutions and integrating with existing systems. This allows healthcare institutions to easily adopt AI technology.

Recent Developments:

• deepcOS® AIR® Launch (Nov 2024): This new platform advancement allows for generating both structured and free-text reports with increased speed and accuracy.

• Osimis Platform Acquisition: This acquisition strengthens deepc's infrastructure for scaling AI use in radiology.

• US Expansion: Partnerships like the one with ImagineSoftware enable wider access to radiology AI for US healthcare practices.

Deepc.ai is a leader in the field of AI-powered healthcare, offering a solution that promises to improve efficiency, accuracy, and ultimately, patient care in the field of radiology.

FDA’s Latest Guidelines for AI Medical Devices

Why these guidelines matter

• AI-enabled medical devices are transforming healthcare, from diagnostics to personalized treatment. However, their iterative nature, driven by real-world data, poses safety and regulatory challenges.

• The FDA’s new Predetermined Change Control Plan (PCCP) streamlines how manufacturers can update these devices, ensuring safety without stifling innovation.

A Quick Look Back

• Previously, any significant update to an AI-enabled medical device required a new FDA submission, often delaying improvements.

• In 2019, the FDA introduced its first proposal to address this issue, emphasizing a “total product lifecycle” approach. This laid the groundwork for the PCCP by acknowledging the dynamic nature of AI and the need for a more flexible regulatory framework.

What’s New?

• Released on December 4, 2024, the guidelines allow manufacturers to predefine planned updates in their initial FDA submission.

• Once approved, changes aligned with the PCCP can be implemented without repeated regulatory approval.

Key highlights of the Guidelines:

• Forward-Looking Planning: PCCPs require manufacturers to outline planned modifications and associated validation processes upfront. This includes defining training data, retraining protocols, and performance evaluation methods.

• Dynamic Regulation: Once a PCCP is approved, manufacturers can implement specified updates without seeking further FDA approval, streamlining the deployment of safer, improved devices.

• Risk Mitigation: The guidance emphasizes rigorous testing and real-world validation to maintain safety across diverse patient populations.

• Transparency: Manufacturers must disclose potential updates in labeling and notify users about significant changes, enhancing trust and usability.

Implementation Process

The PCCP is submitted as part of a marketing application (PMA, 510(k), or De Novo). It includes:

1. Planned Changes: Description of expected modifications.

2. Validation Protocols: Steps to ensure safety and performance.

3. Impact Analysis: Benefits, risks, and mitigation strategies.

Manufacturers can then implement approved updates without submitting additional applications, provided they follow the PCCP.

Future Benefits

• This framework accelerates innovation, enabling faster access to improved AI tools while maintaining oversight.

• For example, AI diagnostic software can quickly integrate new data for enhanced accuracy, benefiting patients without regulatory delays.

The Bottom Line

The FDA’s PCCP guidelines are a significant step toward balancing innovation and safety, ensuring AI devices evolve responsibly while improving patient outcomes.