For telehealth companies like SaluberMD, AI in telemedicine is not optional. Reliable remote diagnostics are essential to delivering quality care at scale. As clinicians use mobile and handheld devices to capture heart and lung sounds and medical images, SaluberMD needed an AI foundation that could operate in real time without compromising compliance or clinician trust.
The problem: unstructured audio and image data slowing clinical decisions in AI in telemedicine
Remote diagnostics workflows generate large volumes of unstructured clinical audio and image data, and in AI in telemedicine, this data varies widely in quality due to device diversity, environmental noise, and differences in how clinicians and patients capture recordings. Inconsistent audio quality can interfere with accurate interpretation of heart and lung sounds, while incorrect or mismatched body-part tagging often requires time-consuming manual validation of diagnostic images. The absence of standardized preprocessing pipelines for medical images and auscultation audio further complicates downstream analysis.
At the same time, telemedicine platforms require scalable, API-driven AI services that integrate seamlessly with existing backends and support low-latency workflows. All of this must operate within strict healthcare compliance requirements, including HIPAA and GDPR, while being deployed on AWS using healthcare-grade security practices. Together, these constraints limit the effectiveness and scalability of AI-driven diagnostics in real-world telemedicine environments.
The solution: agentic AI-driven AWS stack for AI in telemedicine
Delivered as an eight-week pilot by GoML, the solution integrates through secure APIs with SaluberMD’s existing telemedicine platform. Models were trained using PyTorch and deployed on a cost-optimized, healthcare-compliant AWS architecture, creating a scalable foundation for production-grade AI in telemedicine
The system uses EC2 and API Gateway to run FastAPI-based AI inference services, with Bedrock handling image sanity checks. The architecture follows AWS Well-Architected best practices, supporting least-privilege access, encrypted communication via TLS, and secure data handling, with IAM roles replacing static credentials and readiness for future HIPAA-eligible workloads.
1. AI-powered audio enhancement
- Noise reduction and signal isolation for heart and lung sounds
- Clinically optimized WAV outputs suitable for physician review
- Improved diagnostic clarity in mobile and remote AI in telemedicine environments.
2. AI-powered audio classification
- Validation of auscultation audio against labeled body parts such as Heart, Lungs, and Bowel
- Structured JSON outputs with confidence scores to support clinical reliability in AI in telemedicine workflows.
3. Image classification and disease detection
- Body-part classification for Ear, Throat, and Skin
- Disease classification for one selected organ (Skin, nine classes)
- Standardized preprocessing pipelines to support future expansion of AI in telemedicine models across additional organs and diseases.
4. Developer-ready APIs and validation UI
- Secure REST APIs designed for telemedicine backend integration
- Streamlit-based validation UI for clinicians and QA teams supporting AI in telemedicine quality assurance.
The impact of AI in telemedicine at SaluberMD
SaluberMD now operates a production-ready AI foundation that supports scalable, compliant, and low-latency AI in telemedicine workflows. The pilot validated the real-world feasibility of AI-assisted diagnostics in virtual care.
Key outcomes included:
- Clearer diagnostic audio, reducing noise and improving clinical signal quality
- 99% accuracy in auscultation audio classification
- 95.71% accuracy in standardized medical image classification
Before AI vs after AI
“With AI in telemedicine at SaluberMD, we replaced manual validation and inconsistent preprocessing with a reliable, low-latency diagnostic pipeline. Clinicians now get cleaner signals, faster feedback, and greater confidence in virtual care decisions.” - Prashanna Rao, Head of Engineering, GoML
Key takeaways for telemedicine platforms
Common challenges
- Unstructured audio and image data reduces diagnostic reliability
- Manual preprocessing slows clinical workflows
- Lack of standardization limits scalability in virtual care
Practical guidance
- Use agentic AI to automate preprocessing, validation, and inference
- Standardize clinical audio and image inputs before model execution
- Expose AI capabilities through secure, low-latency APIs
- Build on healthcare-grade cloud infrastructure to support compliant, scalable AI in telemedicine
Ready to transform telemedicine with AI?
Partner with GoML to build scalable, compliant, and low-latency AI-powered diagnostic workflows for your healthcare platform.
