Boston University used AI to detect Alzheimer’s in patients up to 6 years in advance

Early detection of Alzheimer’s disease is crucial, as it can prevent or delay the progression of the disease for patients and give caregivers time to prepare. But that’s often easier said than done.

To help increase the chances of earlier detection, researchers at Boston University (BU) created a computer program that uses AI to predict if people would develop Alzheimer’s within six years of showing signs of mild cognitive impairment. Test results, which were published on June 25 in the medical journal Alzheimer’s & Dementia, demonstrate the tool has a 78.5% accuracy rate.

Alzheimer’s disease is among the leading causes of death in the US and was linked to more than 120,000 fatalities in 2022. Physicians often review medical history and use a combination of diagnostic tools like verbal, neurological, blood, and cerebrospinal fluid tests to screen a patient for the disease.

The tool, which has not yet been named, was developed using data from BU’s ongoing Framingham Heart Study that began in 1948. As part of this research, participants take part in neuropsychological tests and interviews. The American Academy of Neurology has recommended neuropsychological assessment to detect dementia since 1996, and this type of testing can both detect and characterize the severity of disease in a patient

There are “tremendous disparities” in who has access to these tests, according to Yannis Paschalidis, professor at BU and director of the Rafik K. Hariri Institute for Computing and Computational Science & Engineering, including financial barriers and geographical challenges if patients don’t live near a medical facility. As a result, many patients are never diagnosed.

Paschalidis and his team used the tool to analyze audio recordings of 166 interviews between 2005 and 2017, finding that 76 participants remained stable while the 90 others experienced a decline in cognitive abilities after six years.

Using speech recognition tools and machine learning to identify correlations between speech, demographics, diagnosis, and disease, the researchers gave each participant a final score that indicated the likelihood their mild cognitive impairment would progress to Alzheimer’s disease. The model pulls content from the interview, like the words spoken and how sentences are structured, instead of using features like enunciation or speed.

Going forward, the researchers plan to study data from not just formal interviews of patients and clinical workers but also everyday speech.

“Even casual speech that one can capture using a smartphone [could] be powerful enough to help us assess and even predict cognitive decline,” Paschalidis said.Boston University used AI to detect Alzheimer’s in patients up to 6 years in advance

FDA Approves First Alzheimer's Drug to Treat Underlying CausePaschalidis added that detecting the disease is crucial to ensure patients get the care they need.

You can't use ChatGPT without knowing the prompts

 

LLM and You can only use ChatGPT if you know the prompts. Here is the guide

ChatGPT Prompting Guide

Here’s What’s Covered :
1. Learning the Terminology: Understand the fundamental terms like Model, Prompt, Input, Output, Token, Max Tokens, Training Data, and Inference.

2. Mastering Commands: Leverage powerful commands to create lists, assume roles, continue conversations, elaborate, summarize, identify gaps, compare, and more.

3. Using Prompt Structures: Implement effective structures such as TREF, SCET, PECRA, GRADE, ROSES, STAR, SOAR, and SMART for better AI responses.

4. Avoiding Common Mistakes: Avoid vague questions, ensure user-centric prompts, simplify language, and maintain clear communication.

5. Utilizing Parameters: Adjust parameters like temperature, diversity_penalty, frequency_penalty, and stop_words to fine-tune responses.

6. Setting the Right Tone: Choose the appropriate tone, whether professional, friendly, enthusiastic, empathetic, instructional, reassuring, inspirational, or formal.

Generative AI refers to a class of artificial intelligence models and techniques that can generate new, original content such as text, images, audio, or even code. Some key characteristics of generative AI include:

- **Creativity**: Generative AI models are trained on large datasets and can use that knowledge to create novel content, going beyond simply retrieving or recombining existing information.

- **Probabilistic Modeling**: Many generative AI models use probabilistic approaches to generate content, predicting the most likely next word, pixel, or audio sample based on the input.

- **Neural Networks**: A common architecture used for generative AI is the neural network, which can learn complex patterns in data and use that to generate new content.

- **Conditional Generation**: Generative AI models can often generate content conditioned on some input, such as generating text based on a prompt or images based on a textual description.

Some prominent examples of generative AI include:

- **Language Models**: Models like GPT-3 that can generate human-like text on a wide range of topics.

- **Text-to-Image Models**: Models like DALL-E that can generate images from textual descriptions.

- **Audio/Music Generation**: Models that can compose original musical pieces or generate speech and other audio.

- **Code Generation**: Models that can write software code based on natural language prompts.

Generative AI has many potential applications, from creative tasks like art and content generation to assistive applications like summarization and query-answering. However, it also raises important questions about ethics, bias, and safety that are actively being studied.

There are some specific Generative AI chats focused on medicine such as DougallMD.

Search engines such as Google, Siri, and others now have Generative AI incorporated in their products.

In fact, ChatGPT has used these search engines to build its database.

Google has become superfluous and replaced with generative AI.

Chat

 Telemedicine Strategies in Rural Hospitals for Emergency Departments and Intensive Care Units

Access Telehealth for Multi-specialties

Rural hospitals present many challenges, recruiting physicians, loss of income, and decreased utilization by the local community. Patients often must travel a hundred miles or more to access health care.  Telehealth provides local access and allows the hospital to retain patients assuring adequate financial resources.  During the past two decades, many smaller facilities have been forced into bankruptcy and/or closed.  Some have merged with larger hospital systems, however, the same challenge remains, obtaining and maintaining adequate physician coverage.

In some cases, physicians will provide coverage on a part-time basis with 'satellite' offices. However, this is not satisfactory since it removes them from their main office.  It also contributes to increased costs of maintaining a second office.

Telehealth can mitigate most of these costs.

The Future of Rural Health Care

VIDEO

Challenges

Ensuring the local community and surrounding regions can rely on their local hospital for a range of specialty care: With limited availability of local specialists, the hospital needed timely access to acute care specialists.

Reducing outbound transfers: The hospital was seeking a solution for reducing the number of patients requiring a transfer to a higher level of care due to a lack of local specialty coverage.

Increasing the case mix index: An increase in the case mix index was a top priority for the hospital as it sought to position itself as a specialty care hub for the region.

Providing a sustainable workload for its outpatient physicians: Many physicians and nurses were covering patients at the hospital and nearby clinic, leading to high levels of burnout and an increased burden on on-site staff.

Attracting Primary Care Physicians is a major undertaking for rural hospitals.  Even more so for specialty physicians.

Population density in much of rural America is very low, many towns have less than 1000 people, barely enough to support one PCP, let alone specialists such as general surgeons, orthopedic surgeons or hospital-based radiologists, and pathologists.

 

The FDA has recently released a new batch of approvals for AI algorithms in healthcare, and the exponential growth continues!

 

The FDA has recently released a new batch of approvals for AI algorithms in healthcare, and the exponential growth continues!

Software designed for medical use bears the burden of safety and accuracy

. The vetting process will be long.

           AI/ML-Enabled Medical Devices

This list contains publicly available information on AI/ML-enabled devices.

Transparency for Machine Learning-Enabled Medical Devices: Guiding Principles

The FDA approval process is an arduous task and adds considerable cost to obtaining FDA approval. Even after approval, the FDA requires updates regarding medical device failure.  The failure rate for ML or AI has yet to be determined.  

Aberrant answers and hallucinations will be assessed carefully.  The FDA will have to assign failure results. Inaccurate results that lead to death or serious injury as a result of AI. ML errors will need to be reported to the FDA much like the FDA reports used for drugs.

The forms are readily available online from the FDA website.