AI in Healthcare: Improving Patient Experience, Diagnosis, and Cost Efficiency

Healthcare providers and technology leaders are under increasing pressure to deliver better patient experiences, reduce costs, and make use of their data—without compromising accuracy or privacy. But deploying artificial intelligence (AI) in healthcare is not just about innovation for its own sake. It’s about solving people-centered problems: How do we make intake more engaging for patients? How do we streamline costly diagnostics without sacrificing precision? And how can we use machine learning to prioritize care and protect patient privacy?

In a recent Synaptiq webinar, Dr. Tim Oates, Co-founder and Chief Data Scientist, explored the real-world applications of AI in healthcare—showcasing how machine learning, computer vision, and large language models are already being used to improve patient experiences, streamline diagnostics, reduce testing costs, and protect patient privacy. Drawing from actual case studies, Dr. Oates offered practical insights into how healthcare organizations can thoughtfully integrate AI to drive both operational efficiency and better clinical outcomes.

Smarter Patient Intake through Adaptive Questionnaires

Challenge: Patients dread filling out long, repetitive intake forms—especially in functional medicine, where intake surveys often contain hundreds of questions. Patients get fatigued, skip questions, or disengage.

Solution: Synaptiq built a dynamic, adaptive questionnaire for our client based on historical patient response data. Instead of asking every question in a fixed order, we used statistical correlations between symptoms to prioritize questions that were most likely relevant for each individual. The questionnaire adapted in real-time based on earlier answers and stayed within topical sections to maintain clarity.

Result: Patients said it “felt like talking to a doctor.” More relevant questions were answered earlier, improving data quality and enhancing the overall intake experience—all without using advanced deep learning, just thoughtful application of statistical pattern recognition.

Automating Ultrasound Annotation for Cardiovascular Health

Challenge: Measuring intima-media thickness (IMT) via ultrasound is an excellent indicator of cardiovascular health, but it's labor-intensive. Skilled technicians must manually identify key artery boundaries across video frames.

Solution: For our client, Synaptiq developed a neural network–based segmentation model to automate the detection of carotid artery structures in ultrasound videos. Despite limited training data, we designed the system to match the precision of human measurements and visualize plaque buildup or artery thickening over time.

Result: Clinicians can offer patients a clear visual explanation of their cardiovascular risks, increasing treatment adherence. Automated annotation reduces technician workload and improves consistency in diagnosis.

Reducing the Cost of Diagnostic Testing 

Challenge: VO₂ max, a key metric of cardiovascular fitness, usually requires a costly lab setup (a metabolic cart) and in-person testing. This limits scale.

Solution: We helped a client develop a lower-cost alternative using stepper exercise and smartwatch data. By training a regression model on paired high-fidelity and low-fidelity data, we predicted VO₂ max using only wearable data and simple movements.

Result: Patients received personalized “exercise prescriptions” to improve health without needing lab visits, making preventive care more accessible and affordable.

The Role (and Limits) of Large Language Models in Healthcare

Large language models (LLMs) like ChatGPT can support clinicians in non-patient-facing tasks:

• Summarizing medical documents
  
  
• Drafting doctor's notes in different tones
  
  
• Enhancing spreadsheet analysis (e.g., using Gemini in Google Sheets)

The challenges with LLMs include:

• LLMs are not trained clinicians and must be used under human supervision
  
  
• Integration into clinical workflows is much harder than casual one-off use
  
  
• Bias, security, and explainability remain significant concerns

As a best practice, keep humans in the loop. Use LLMs to augment—not replace—clinical decision-making. When properly deployed, they can reduce administrative burden and free up human time for empathy and judgment.

Conclusion

From adaptive patient intake to automated ultrasound analysis, healthcare organizations can implement AI in ways that improve outcomes, protect privacy, and lower costs. The projects we shared above span traditional machine learning, computer vision, and large language models—but the common thread is always a focus on real people and practical impact.

Want to explore how AI can reduce friction and improve outcomes in your healthcare organization? Contact Us to start the conversation.