EIDA 1
2
3
4
5
6
7
8
9

E.I.D.A

Enhanced Intelligent Diagnostic Assistant [Best Startup, Best Overall, Best UI/UX (speech interaction)]

Inspiration

Hospitals across the world are struggling from a lack of staffing as the need for healthcare continues to grow. These circumstances mean hospitals are understaffed and healthcare professionals are overworked, meaning the issues patients face may be misdiagnosed or not adequately addressed. These circumstances are not good for anybody. For the patients, these mistakes can be life threatening or even be fatal. For the hospital and staff, this means increased pressure to perform and repercussions from making mistakes under pressure.

What it does

We present E.I.D.A., which stands for Enhanced Intelligent Diagnostic Assistance. E.I.D.A. is a patient preprocessing tool designed to reduce the workloads of primary healthcare providers while still ensuring that patient concerns and needs are adequately heard. By conversing with E.I.D.A through speech, the patient can provide details about their ailments. E.I.D.A gathers this information and compiles a history of present illness (HPI) using a large language model to be read by a healthcare provider when they are about to meet the patient.

Furthermore, E.I.D.A. can respond dynamically and ask qualifying questions regarding a patient’s situation to further elaborate on their present condition. If a patient makes a comment where clarification is needed, E.I.D.A. is capable of asking the patient to clarify or provide more details, similarly to how a real healthcare provider would interview patients. Such a product will enable primary care providers to better isolate issues and address them while reducing workloads.

E.I.D.A. is capable of speaking in a variety of languages to better address the needs of patients and allowing them to speak in a language they are comfortable with. This means that primary healthcare professionals can get the important details of a patient's condition while overcoming language and cultural barriers that traditional technologies fail to address.

How we built it

E.I.D.A. is powered by OpenAI’s GPT-3.5 large language model. Patients interface with E.I.D.A. using speech and natural language through the usage of the cognitive services from Microsoft Azure. Our usage of GPT-3.5 is controlled through Guidance, a recently released tool by Microsoft to guide the text generation capabilities of large language models like GPT-3.5.

Challenges we ran into

Some challenges we ran into were making use of Guidance, the software released by Microsoft. Since this is a tool released about a month ago, there were not many working examples to use as inspiration. We found difficulty in processing the outputs of the model and making use of the full capabilities of Guidance.

This also limited the scope for our demonstration as we originally planned for several multi-language support to be implemented in E.I.D.A. However due to shortage of time we limited our demonstration to English, Hindi, Chinese and Japanese. We feel that integrating support for various languages can help make healthcare accessible to a much wider audience, especially to the communities who do not share the same language as those who are providing the primary care. Additionally, we discovered that with better prompt engineering we can utilize the GPT-3.5 model more effectively recognize different languages and have the Azure Cognitive Speech module translate the language changes.

Accomplishments that we're proud of

We’re proud of the fact that we are able to have E.I.D.A. ask patients for qualifying questions about concerns or symptoms that they might have. Since patients may typically considered an unreliable narrator, this is a crucial feature to have to be able to pull accurate information from a patient. Furthermore, E.I.D.A.’s questions remain short and concise so that the patient is not overwhelmed or bombarded by questions. It fast, efficient, and accurate in making notes about the details the patient outlines.

What we learned

We found that while the use of large language models makes certain aspects of this project easier, it also brought new challenges. For example, testing the integration of the large language model was difficult as it was not necessarily guaranteed that the model would return the same text, even for the same prompt. This meant that we could not build automated test cases and had to test changes each time manually.

What's next for E.I.D.A

A feature we would like to implement in E.I.D.A is language recognition, as it would help make healthcare to many more groups of people. We are also hoping to integrate object recognition capabilities to analyze the emotions of a patient similarly to how humans are able to infer emotion from someone’s face.

Business potential

The product incorporating speech-to-text technology and language detection has substantial business potential within the healthcare industry. Here are some key aspects:

Improved Efficiency
Enhanced Patient Experience
Market Demand
Accessibility and Inclusivity
Competitive Advantage Considering these factors E.I.D.A presents opportunities for revenue generation through software licensing, subscription models, or partnerships with healthcare organizations. It has the potential to improve operational efficiency, patient satisfaction, and ultimately contribute to better healthcare outcomes.

Impact

The impact of the hack and the number of people it will help depend on several factors, including the scale of implementation, user adoption, and the specific context in which it is applied. However, let's consider the potential impact and beneficiaries in a general healthcare context:

Time-saving
Multilingual Support
Improved Accuracy
Accessibility for Individuals with Disabilities
Larger Selection of Patients The actual number of people who will benefit from E.I.D.A will depend on the healthcare provider, especially in healthcare settings that serve multilingual populations or prioritize accessibility for individuals with disabilities. Overall, the E.I.D.A has the potential to make healthcare more efficient, accessible, and inclusive, benefiting both healthcare providers and patients.