🚨 Pulse of Innovation 🚨

Breaking news in the healthcare AI

Generative AI-powered Robots in Healthcare: Are They the Future?

At the European Respiratory Society (ERS) Congress 2024, attendees were introduced to AMECA, a humanoid robot powered by chatGPT 4.0. The session explored AI's potential to transform respiratory care, with experts discussing its benefits and challenges.

Humanoid or Non-humanoid Robots: The Next Frontier

Both Humanoid and non-humanoid robots have many potential applications in healthcare like AMECA can communicate with patients using natural language, simulating human conversation. This could revolutionize patient interactions but also raise questions about their limitations and the need for careful integration into clinical practice. Their efficiency can be enhanced by the use of advanced generative AI like GPT 4.0

AMECA: A Case Study

AMECA, powered by ChatGPT 4.0, demonstrated its ability to answer basic questions about respiratory conditions. However, it struggled with more complex queries, highlighting the need for further understanding of context and nuance.

Question: Can you diagnose chronic obstructive pulmonary disease (COPD) based on forced expiratory volume in 1 second (FEV1) alone?

AMECA Answer: A diagnosis of COPD cannot be based on FEV1 alone. It requires a combination of spirometry results, including a FEV1/FVC [forced vital capacity] ratio along with clinical symptoms and a medical history.

Limitations of AMECA:

• The answers were to the point and not inaccurate, but they seemed less human, particularly in their length and lack of human-centric conversational elaboration, which is often the case between a patient and a doctor.

• These robots also struggled to answer and handle more complex questions when their answerers were usually short and merely one sentence.

Better Generative AI Control of Robots

• One of the biggest challenges in applying machine learning to robotics is the paucity of data. While computer vision and natural language processing can piggyback on the vast quantities of image and text data found on the Internet, collecting robot data is costly and time-consuming.

• An LLM that can be integrated into all types of Robots can increase the efficiency of a machine-learning model by improving the data and can potentially be used in humanoid and non-humanoid robotic applications.

• CrossFormer is a single AI robotic platform that can be used to control different types of robots in the future. This article shows that CrossFormer performed better than standalone applications in different robots.

• Such applications of humanoid robots will be really beneficial in healthcare, enabling better patient interaction, physician aid, and more human-centric data-driven decision-making.

🧑🏼‍🔬 Bench to Bedside👨🏽‍🔬

Developments in healthcare AI research and innovations

Large Language Models based AI tools for our Research Data Collection: Coming soon….

• Real-world observational data-driven and outcomes-based research studies have become increasingly prominent in recent decades. Although evidence of randomized clinical trials is considered superior, these large real-world studies have often identified critical deficiencies and knowledge gaps.

• The most cumbersome aspect of real-world observational studies is computing a structured database that can answer the research questions. Often, the clinical data required is free text that needs to be converted to structured, analyzable data. With major improvements in natural language processing (NLP) models, more of these studies could be feasible.

• A recent study published in Nature Digital used the LLAMA-2 NLP on 500 patients to identify quantitative information from clinical text and evaluate its performance in identifying features of decompensated liver cirrhosis.

• One key feature required from these NLP models is the preservation of privacy, which is often a risk when allowing the extraction of data from patient records.

• The study demonstrated the preservation of absolute patient privacy, with cirrhosis detection sensitivity at 100% and specificity at 96%, thus achieving greater success than smaller LLMs.

• More such studies will open doorways for LLM-based data-driven outcomes research studies, which can answer many gaps in the evidence.

🧑🏽‍⚕️ AI in Clinic 🏥

Developments in healthcare AI research and innovations

AI - Jargon Buster

Automation/Robotic Process Automation (RPA)

The use of specialized software and technology to carry out repetitive tasks, following a set of instructions and workflows by humans. These tasks typically remain consistent over time, including sending appointment reminders, missed appointment notifications, or even receipts after online purchases. If a task is not explicitly outlined in the instructions, the machine cannot perform it. 

Algorithm

An AI system uses a set of well-defined rules or processes to conduct tasks such as discovering new insights, identifying patterns, predicting outcomes, and solving problems.

Artificial intelligence (AI)

AI is the capability of a computer system to mimic human cognitive functions such as learning, problem-solving, interpreting visual information, understanding, and responding to spoken or written language. It uses math, logic, and patterns learned from data to simulate human reasoning and make decisions and recommendations. 

Machine learning (ML)

ML is a subset of AI that enables machines to learn and improve from experience without explicit programming automatically. By using set processes to analyze large amounts of data, ML systems can identify patterns, help make decisions, and improve their performance with little to no human intervention. 

Prompt (engineering)

A prompt is a question, command, or statement input into an AI model to initiate a response or action. It facilitates interaction between a human and the AI to generate the intended output.

🤖 Patient First, AI Second🤖

Ethical and Regulatory Landscape of Healthcare AI

Humanity's Last Exam

The Center for AI Safety (CAIS) and Scale AI have launched "Humanity's Last Exam," an ambitious project to create the world's most challenging AI benchmark. As reported by Reuters, this initiative comes in response to rapid advancements in AI capabilities, with recent models like OpenAI's o1 "destroying the most popular reasoning benchmarks." The exam seeks to gather expert-level questions across various fields, offering contributors co-authorship opportunities and substantial prizes to measure how close AI systems are to achieving human expert-level capabilities.

The Humanity's Last Exam project aims to create a relevant benchmark as AI capabilities rapidly advance. By setting a new bar for AI assessment, the exam seeks to influence both market leaders and startups in their AI research and development efforts, potentially driving significant investments in the field.

The exam will include 1,000 crowd-sourced questions due by November 1, 2024. These submissions will undergo a rigorous peer review to ensure quality and relevance. To maintain the integrity of the benchmark, a subset of questions will be kept private. This approach helps

prevent AI systems from simply memorizing answers and allow for a more accurate assessment of their true capabilities.

A $500,000 prize pool has been allocated, with the top 50 questions earning $5,000 each and the next 500 questions receiving $500 each. Beyond monetary incentives, successful submissions will earn their creator’s co-authorship on the resulting paper, providing recognition in the academic and AI research communities. This collaborative approach has already attracted submissions from researchers at prestigious institutions such as MIT, UC Berkeley, and Stanford.