top of page
  • Sameer Shaikh

Newsletter from The Neural Medwork: Issue 1

Updated: Dec 13, 2023


 

Abstract:


Welcome to the first edition of The Neural Medwork, your specialized guide to the world of AI in healthcare. We understand that as healthcare professionals, your time is precious, and staying updated with the rapid advancements in AI can be daunting. Our newsletter is designed to make this journey easier and more accessible, providing you with concise, relevant, and practical insights into AI in healthcare.


In every issue, we will cover three main sections: one AI Concept, one relevant research paper, and some practical tips & tricks on utilizing available AI tools. These segments are carefully curated to build your foundational knowledge of AI, keep you informed about the latest research, and offer practical advice on utilizing AI tools at the bedside.


 

AI Concept: Neural Networks

What better place to start than the concept from which our newsletter takes its name. A neural network in AI can be thought of as the brain of a physician. Just as a doctor gathers a wealth of information from a patient’s history, physical examination, and diagnostic tests before making a diagnosis or treatment plan, a neural network processes a wide array of inputs to reach a conclusion.


Consider a neural network as a physician’s brain engineered by computer scientists or AI engineers. It absorbs various inputs, much like you do when seeing a patient. This information passes through the neural network layer by layer, where it is interpreted, digested, and analyzed before a final output is generated.


Neural networks can be unimodal or multimodal, handling different types of data, similar to how physicians integrate diverse clinical insights. Engineers train these models for specific tasks, akin to how clinicians develop expertise throughout their training. As more data and guidance are provided, the network’s ability to produce meaningful outputs and make decisions improves - mirroring a clinician’s growth with training and experience.


In AI, neural networks serve as a key component for most products. From predicting future events and interpreting medical images to devising treatment plans, most AI tools have their own neural network that is trained and tested on the relevant data necessary to complete a task. Key considerations when evaluating an AI solution include understanding how its neural network was trained, the data it was fed, and how it reaches its conclusions. Much like a clinician’s intuition, sometimes the workings of a neural network might not be entirely transparent. However, in cases where this process can be understood, this transparency - understanding how decisions are made - is crucial.


In the realm of AI, there are various types of neural networks that engineers and data scientists utilize, especially in healthcare applications. Each type has its unique architecture and is suited for different tasks. Here are three of the most common neural networks you will see in healthcare applications:

  • Convolutional Neural Networks (CNNs):

    • Specialized for processing data with a grid-like topology, such as images.

    • CNNs excel in medical imaging tasks, like identifying abnormalities in X-rays or MRIs, due to their ability to capture spatial hierarchies in data.

  • Recurrent Neural Networks (RNNs):

    • Designed to handle sequential data, such as time series or speech.

    • In healthcare, RNNs are useful for analyzing patient data over time, predicting disease progression, or processing speech in patient-physician interactions (i.e. Ambient AI solutions).

  • Deep Neural Networks (DNNs):

    • Composed of multiple layers, enabling the model to learn complex patterns.

    • DNNs are versatile and can be applied to a variety of tasks in healthcare, from patient data analysis to aiding in diagnosis and treatment plans.

Each of these neural networks brings a unique approach to analyzing data, making them indispensable tools in the AI healthcare toolkit. As we continue to explore AI, understanding these networks will deepen your appreciation of how AI technologies are tailored to specific healthcare challenges.


Neural Network Architecture
Neural Network Architecture

















 

Relevant Research Paper


Title: Competencies for the Use of Artificial Intelligence–Based Tools by Health Care Professionals


Purpose of the Study:

  • To define AI-related clinical competencies for healthcare professionals due to the expanding use of AI in clinical tools.

  • This qualitative study involved expert interviews to identify these competencies.

Methodology:

  • In 2021, a multidisciplinary team conducted interviews with 15 experts in the use of AI-based tools in healthcare.

  • The interviews were semistructured, focusing on necessary clinical competencies.

  • The transcripts were coded and thematically analyzed to develop competency statements.

Key Findings:

  • Six competency domain statements and 25 subcompetencies were identified.

  • The competency domains include:

    1. Basic Knowledge of AI: Understanding of AI and its healthcare applications.

    2. Social and Ethical Implications of AI: Awareness of how socio-economic systems influence AI tools and their impact on ethics and equity.

    3. AI-Enhanced Clinical Encounters: Ability to integrate diverse information sources in AI-enhanced patient care.

    4. Evidence-Based Evaluation of AI-Based Tools: Skills to evaluate AI tools’ quality, accuracy, safety, biases, and appropriateness in clinical settings.

    5. Workflow Analysis for AI-Based Tools: Adaptation to changes in team roles and workflows due to AI tool implementation.

    6. Practice-Based Learning and Improvement Regarding AI-Based Tools: Engagement in professional development and improvement activities related to AI in healthcare.

Conclusion:

  • These identified competencies can guide future teaching and learning programs to maximize AI-based tools' benefits and minimize potential harms in healthcare. We hope to cover knowledge within all of these areas to give clinicians a strong foundation in AI-related clinical competencies.


Table for AI-related clinical competencies
Table for AI-related clinical competencies

Russell, Regina G. PhD, MA, MEd1; Lovett Novak, Laurie PhD2; Patel, Mehool MD3; Garvey, Kim V. PhD, MS, MLIS4; Craig, Kelly Jean Thomas PhD5; Jackson, Gretchen P. MD, PhD6; Moore, Don PhD7; Miller, Bonnie M. MD, MMHC8. Competencies for the Use of Artificial Intelligence–Based Tools by Health Care Professionals. Academic Medicine 98(3):p 348-356, March 2023.



 

Tips and Tricks: Prompting ChatGPT


As healthcare professionals, you'll likely find ChatGPT to be one of the most accessible and versatile AI tools due to its widespread availability. The key to effectively utilizing ChatGPT lies in the art of prompting, which is referred to as prompt engineering. Properly framing your questions and inputs is crucial for obtaining relevant and useful responses. Over the next few weeks we will highlight a number of tips and tricks to master the art of prompt engineering so that you can make the most of ChatGPT. A simple way to start is by looking at a mnemonic that Allie Miller, a prominent voice on AI in LinkedIn, coined as WISER:


W - Who is it: Start by defining an identity for ChatGPT, such as "You are a medical research analyst." This sets the stage for the type of information and perspective you're seeking.

I - Instructions: Be clear about what you want ChatGPT to do. For example, instruct it to explain a medical procedure in layman's terms. The clearer your instructions, the more accurate the response will be.

S - Subtasks: Break down your query into smaller, more manageable tasks. If you're asking about a disease, you might start by asking for a definition, then its symptoms, followed by treatment options. This step-by-step approach helps ChatGPT provide structured and comprehensive information.

E - Examples: ChatGPT excels at following examples or templates. If you want a specific format or style, like a patient information leaflet, provide a similar example for it to mimic.

R - Review: Reflect on the response you get. If it's not quite right, don't hesitate to refine your prompt, ask follow-up questions, or request more detailed explanations. The interactive nature of ChatGPT means you can continually refine your query to get closer to the desired answer.


By applying the WISER approach, you can effectively tailor ChatGPT's vast capabilities to your specific medical and informational needs, enhancing both your practice and patient care.


We hope you enjoyed the inaugural edition of The Neural Medwork. Stay tuned for our weekly newsletter, and exciting Vlog’s with prominent experts in the field of AI and healthcare. If you haven’t already, check out our first Vlog on How to create GPT with no code right here:




Thanks for tuning in


Sameer & Michael

Comments


bottom of page