🚨 Pulse of Innovation 🚨

Breaking news in the healthcare AI

ChatGPT 4o, how can we in healthcare make the most out of it?

• GPT-4o, a natively multimodal AI marvel, can process text, audio, and images simultaneously and even talk. This new model promises a more natural human-computer interaction, like talking to a knowledgeable friend or personal assistant. GPT-4o is 2x faster and available to everyone(without a subscription). You can use it for all your generative AI needs, including customer service bots, content creation, personalized education tools, software development, and more.

• Goal: Over the next few issues, we will showcase multiple cases of GPT4o being used in clinical practice. For today, see this very simple use case of language translation. As a physician at one of the largest county hospitals, I see patients who speak multiple languages; I have used this in the clinic to communicate with them, and it has been very intuitive and easy.

   

🧑🏼‍🔬 Bench to Bedside👨🏽‍🔬

Developments in healthcare AI research and innovations

StrokeClassifier: The deep-learning model to classify stroke

• Classification of acute ischemic stroke is key to secondary prevention strategies. Two academic hospitals built a deep-learning model using electronic health records and generated various classifications for stroke.

  • High Accuracy and Utility: StrokeClassifier demonstrates robust accuracy (0.744 cross-validation, 0.70 MIMIC-III) and a high weighted F1 score (0.74 cross-validation, 0.71 MIMIC-III), indicating reliable performance in classifying ischemic stroke etiologies.

  • Clinical Decision Support: The tool provides valuable support for clinicians by aiding in the accurate and timely diagnosis of stroke etiology, which can enhance patient care and streamline the initiation of targeted prevention therapies.

  • Feature Utilization: Using EHR data and NLP techniques to extract and utilize top predictive features (e.g., atrial fibrillation, age) enhances the model's relevance and applicability in clinical settings.

• Variability and Generalizability: The variability in documentation practices and testing protocols between different sites limits the generalizability of the model's performance, necessitating further validation in more diverse cohorts. Cryptogenic Diagnosis Reduction: While the certainty heuristic reduces cryptogenic diagnoses, the residual cryptogenic rate (7.2%) still indicates a need for further refinement to minimize uncertainty in stroke etiology classification.

👩‍⚕️AI in the Clinic 🏥

Real-world and practical use of healthcare AI in clinic

At-home patient monitoring for early detection of Heart Failure

• Peer Bridge Health recently reported data from its clinical trial which used its Cor-3. This 2-lead EKG technology does home monitoring and uses AI to determine whether the patient is in heart failure.

• The clinical trial data reported at the Hearth Rhythm Conference showed a 96% accuracy in correctly detecting patients with heart failure. Being a low-cost remote monitoring system, this could be an important game changer in the field of heart failure remote monitoring of patients.

🩺 Start-Up Stethoscope 💵 

Trending startups and technologies impacting clinical practice

Dot Compliance - ready-to-use Quality Management System

• Dot Compliance offers ready-to-use Quality Management Systems (QMS) specifically designed for the life sciences industry. Their solutions streamline quality and compliance processes using AI-powered tools and cover many needs, from document management to audit management. The company announced it closed $17.5 million in Series B extension funding, bringing its total funding to $50 million. Their products include QMS Xpress for foundational processes, Compliance Xpand for expanded functionalities, and Enterprise Xact for tailored solutions. Key features include rapid deployment, cost-effectiveness, and extensive process coverage.

• Dot Compliance Technology's limitations include Integration into your existing system and its dependence on Salesforce. While offering pre-configured solutions, some organizations may still need customizations to meet specific requirements, potentially adding to implementation time and cost.

🤖Patients First, AI Second🤖

Ethical and regulatory landscape of healthcare AI

Is your Large Language Model Foundation Model Transparency Index?

• Context. Foundation models like GPT-4 and Llama 2 are used by millions of people. While the societal impact of these models is rising, transparency is on the decline. If this trend continues, foundation models could become just as opaque as social media platforms and other previous technologies, replicating their failure modes.

• Design. The Center for Research on Foundation Models introduces the Foundation Model Transparency Index to assess the transparency of foundation model developers. We designed the Index around 100 transparency indicators, which codify transparency for foundation models, the resources required to build them, and their use in the AI supply chain.

• Execution. For the 2023 Index, the model scores 10 leading developers against our 100 indicators. This provides a snapshot of transparency across the AI ecosystem. All developers have significant room for improvement that we will aim to track in the future versions of the Index.