KEYWORDS: ChatGPT, Healthcare, ChatGPT Healthcare, GenerativeAI, UseCases, HealthcareAI, Healthcare multimodal data, healthcare data.
Last Wednesday, Amazon announced a generative AI-based clinical documentation service, called HealthScribe, which allows providers to build clinical applications that use speech recognition and generative AI to create transcripts of patient visits, identify key details, and create summaries that can be entered into an electronic health record. [1]
Generative AI is rapidly transforming the healthcare industry.
In the previous article of this series, we covered topics such as the potential benefits of ChatGPT to transform the healthcare industry and ChatGPT integrated use cases.
For more information, please read:
How ChatGPT and Generative AI can transform Healthcare (Part.1)
In this article, we will discuss the following topics:
1. Existing challenges of ChatGPT in Healthcare;
2. The importance of data quality and quantity;
3. New opportunities to explore the value of multimodal healthcare data;
4. The strengths of maadaa.ai in the healthcare industry.
1. Existing Challenges of ChatGPT in Healthcare
With the introduction of ChatGPT and other advanced AI, we have reached a turning point in artificial intelligence.
Despite the great potential of ChatGPT, there are still some issues that cannot be ignored.
Among them, the most important challenges are accuracy and ethical issues.
ChatGPT’s data ends at September of 2021 and always shows the lack and errors in certain fields such as finance, healthcare, etc.[2]
As we know, the key to the success of AI models like ChatGPT is the quality and scope of the datasets used in their training, which allows for a more comprehensive understanding and more accurate responses.
However, the outdated data and insufficient quantity and quality of data in specialized areas suggest inaccuracies in ChatGPT in healthcare which will lead to serious consequences.
In addition, ethical considerations are another major challenge of ChatGPT in healthcare, which include:
- Bias in data. Based on large amounts of training data, ChatGPT may contain biases. This can have a negative impact on patient care. Ensure that the data used to train the model is diverse and representative of the population it will serve.
- Blind trust in ChatGPT. It is natural for users to blindly trust the system to be 100% accurate at all times. ChatGPT should be used as an adjunct or underlying structure, not a replacement, and all work and information it provides should be verified.
- Lack of regulation. It is important to ensure that the use of ChatGPT complies with all relevant healthcare laws and regulations.
- Privacy and security. AI will need to feed data, which will often be confidential patient information. Healthcare systems must be set up to ensure that patient information remains private.
(Photo by National Cancer Institute on Unsplash)
2. AI in Healthcare: The Significance of Data
Recently, the Senior Vice Dean of the Perelman School of Medicine sat down with ChatGPT to ask how it could be useful in healthcare.
When asked, “What will it take before ChatGPT can be applied in everyday health care?” ChatGPT responded, “To ensure ChatGPT can provide accurate and relevant responses, it will require access to high-quality, diverse and large datasets, especially in the field of healthcare.”[3]
In addition, here is another quote from an article titled “What Can ChatGPT Do for Healthcare Practices?” that says, “Comprehensive datasets are essential for enhancing the performance of AI models, as they enable them to understand more comprehensively and generate more accurate responses.”[4]
Therefore, combining the existing challenges and the significance of AI training data in healthcare, there are two issues that need to be addressed.
1. How to handle the value of multimodal healthcare data?
2. The importance of collecting large amounts of data and providing professional annotation?
We will discuss them in the following section.
2.1 How to handle the value of multimodal healthcare data?
In fact, it is relatively easy for a well-designed machine-learning model to detect a lesion or grade on a particular cell.
However, one of the remaining challenges is to find comprehensive and patient-specific ways, such as predicting a patient’s response to treatment or prognosis and early intervention.
Therefore, analyzing a combination of data in different forms (integrating multimodal data) can significantly help physicians to have the most comprehensive understanding of the patient, their background, and the evolution of the disease.
A study named “Multimodal Biomedical AI” proves that organizing higher quality image-text datasets may be more important than generating large unimodal datasets and other aspects of algorithm development and training, according to DeepMind [5].
The availability of multimodal data in AI training can help achieve better diagnostic performance in a number of different tasks.
For example, recent work has shown that the combination of imaging and EHR data outperforms either modality alone for detecting pulmonary embolism and distinguishing between common causes of acute respiratory failure, such as heart failure, pneumonia, or chronic obstructive pulmonary disease[6].
Drug discovery is also an important example. Many tasks could benefit from multidimensional data, including target identification and validation, prediction of drug interactions, and prediction of side effects[7].
It follows that the primary requirement for the successful development of multimodal data-based applications is the collection, curation, and harmonization of well-phenotyped and large annotated datasets.
2.2 The importance of collecting large amounts of data and providing professional annotation?
In fact, today we are far less capable of organizing and storing such data than we are of analyzing it.
To have the ability to capture multidimensional healthcare data, it is necessary to find a reliable AI data work partner to help you across industries and sectors is needed to collect and link large and diverse multimodal health data.
Here are some considerations for working with AI training data for healthcare.
2.2.1 Is the training data legally compliant?
With its professional on-site data collection program management, maadaa.ai ensures full compliance with local data policies with guaranteed data copyright and scenario-based data set customization.
maadaa.ai has achieved ISO/IEC 27001 Information Security Management System certification, the world’s most recognized standard for Information Security Management Systems (ISMS).
maadaa.ai has more than 200 collaborations with leading technology companies, universities and institutions.
2.2.2 Can the quantity and quality of the collected data meet the requirements?
maadaa.ai, founded in 2015, is a comprehensive AI data service company supplying the AI industry with professional data services in text, voice, image, and video data types.
To support different types of clients, maadaa.ai has established a global data collection network in more than 40 countries.
maadaa.ai has participated in many data projects in the medical field, accumulating a lot of experience, resources, and a large number of experts in various fields, as well as frontline medical practitioners, including clinicians, pharmacists, nurses, healthcare professionals, etc.
AI experts from maadaa.ai will analyze your application scenario. Reference cases will be provided as appropriate. We will then work with you on your sample dataset to ensure the project parameters meet your requirements.
Our AI data engineers will develop tools applicable to your problem sets, while our operations team will manage production through an online platform and a dedicated on-site data factory.
2.2.3 Can annotators reach healthcare industry standards?
With years of annotation experience, maadaa.ai’s annotation tool, annotation process, and project staff management have been verified by more than 200,000 tasks to be efficient and easy to use.
maadaa.ai recruits, tests and trains the right annotators according to customer needs and has years of experience working with professional annotation teams.
Our in-house developed annotation tools support customized design, development, and deployment to ensure service scalability. maadaa.ai is able to support over 10000 personnel annotation at the same time, and service TPS over 10000.
Flexible configuration quickly adapts to a variety of data annotation demands, millisecond response export.
2.2.4 How to prevent data leakage?
maadaa.ai provides multiple security protection, enterprise intranet deployment, low-level authority control, and customized data access security path.
Based on our successful practices of thousands of commercial AI data projects, maadaa.ai launched the industry-leading MaidX data platform, providing one-stop AI data solutions for industrial AI product development.
In fact, maadaa.ai has worked in specific medical fields such as cellular data annotation projects, brain segmentation data annotation projects, retinal data annotation projects, and other medical data annotation projects.
To find out more information, please read:
maadaa.ai’s motivation is to help AI industry customers efficiently collect, process and manage data, and perform model training for fast and cost-effective AI technology adopt
Reference List:
[1] https://www.healthcaredive.com/news/amazon-generative-ai-clinical-documentation-healthscribe/688996/
[2] https://maadaa.ai/blog/chatgpt-for-enterprise-scenarios-how-to-cope-with-the-data-challenges/
[3] https://catalyst.nejm.org/doi/full/10.1056/CAT.23.0043
[4] https://www.entrepreneur.com/science-technology/what-can-chatgpt-do-for-healthcare-practices/450626
[5] Hendricks, L. A., Mellor, J., Schneider, R., Alayrac, J.-B. & Nematzadeh, A. Decoupling the role of data, attention, and losses in multimodal transformers. Trans. Assoc. Comput. Linguist. 9, 570–585 (2021).
[6] https://www.sciencedirect.com/science/article/pii/S1359644622001568
