Building our AI for the Wound Capture Process

post by rapid-healthcare on January 14th, 2021

As team Rapid expands on its roadmap for 2021 we wanted to let you know about our new product launch this year. Wound Capture AI is the solution for managing the wound documentation process to prevent untrained human diagnostic errors  as well as a cumbersome current workflow. Wound Capture AI is designed to be an all in one event record for all the details of a patients wound care treatment, early detection and progress. The application’s base mechanic will be an image capture, wound sizing & recognition function that will be a mobile first wound documentation solution.

Current Problems in Wound Documentation Process:

  • Cumbersome workflow (100% chief complaint from nurses interviewed). Tasks involving multiple departments creating unnecessary complexity using Photo-Print-Paste (PPP) methods.
  • More than 70+ observations performed on 20 inpatient units. Averaged 30 min. to perform the P-P-P workflow!
  • Equipment failure (digital camera, SD card, or photo printer).
  • Photos taken were not HIPPA secured.
  • Photos were scanned and uploaded after patient was discharged.
  • Delayed preventive or initiating wound treatments.
  • Difficulty for wound consultants to triage wound urgencies.
  • Delaying physician communication.
  • Risk for wound related sepsis and secondary complications or infections.

The native camera API is utilized to implement the image capture mechanic. On capture of a wound image, the image’s data is sent to the image recognition process to be analyzed by our training models. Once processed, the user is given graphical and text feedback based on the confidence scores returned by the model, along with ‘next step’ directions, in the user interface if detected for a particular problem.

The mobile app prompts the user to login via badge id and pin and then scan the patient’s ID band for positive patient identification and validation at the point of care. The app prompts the provider to identify the body location of the wound and when the image of the wound is selected, wound capture will in real-time give predictability of wound progress, calculate dimensions of width and length using AR, make notes, review previous images and then help to compile a potential treatment record once tied with patient information, providers name, identification number, wound image description, measurements as well as real date and time stamp.

This image is stored as a PDF once captured via the app and can be sent via email to any authorized provider of care and/or uploaded to the patient’s medical record via FHIR connection to any EHR/EMR system. Each event is catalogued in Rapid’s H-gateway web Portal where we store our machine learning models. There is a recorded history that can be reviewed, updated via previous images captured for that patient and at any time via a mobile device used for machine inference of the success of the wound treatment.

Wound Capture solves the problematic areas of wound management in the following ways:

    • To provide thorough, progressive and subjective wound assessments.
    • To identify early infection of the wound or surrounding tissues, with quality images, size calculations and provider notes.
    • To provide enterprise access whether used within inpatient or outpatient settings.
    • To diagnose an injury due to a misinterpretation of symptoms or inadequate examination
    • Photos are HIPPA compliant, pictures are only saved on EHR, not on the mobile device.
    • Location documentation ability to add notes and create PDF document with embedded photos.
    • 98% increase in nursing satisfaction, excitement and approval.
    • Minimize task related burnout.
    • Improvement in Wound Department response time: 1) By triaging wound consults and ordering
    • Medical records department – can minimize technicians time who is typically doing manual scan
    • How we defer from others: is that we do digital measurement during the capture process.
    • AI model for learning types of wounds and image recognition based on digital documentation of wounds.
    • Typical workflow that takes 20-60 minutes to achieve, can take just a few minutes with immediate mobile capture & upload and organization into patient’s medical record.and upload of each wound pictures into patient’s remotely.
    • Can minimize interdepartmental traveling time.
    • Decrease risk of septic wounds due to early treatment and detection by looking at previous photos.
    • Increase insurance reimbursement.
    • A conservative labor saving estimated at $600,000/year in labor costs.

On-device Inference 

For this experience in particular, we decided to use an on-device (over a cloud-based) image recognition approach. On-device inference has a number of advantages for this application: 

  • No network dependency. Storage of the model on the user’s mobile device ensures that the application will work in the absence of an internet connection 
  • Low-latency. On-device inference yields a snappier user experience, due to not having to communicate with the cloud/server all the time (only to sync).
  • Users will be able to receive real-time feedback from the model in the application’s camera view
  • Privacy It helps to circumvent privacy issues, as there are no images being sent to servers over the internet. 
  • Cost-effective to scale Using an on-device approach over a Cloud based ML service is advantageous in scale, since cloud-based services are monetarily expensive and could prohibit future scale should the app ever gets heavy usage. 

On-device image recognition is implemented in the app. Once the model has been trained, the learned model weights are “Quantized” from 64-bit floating point arithmetic to 8-bit integer operations that are more efficient to compute on a mobile device.

Wound Capture AI is currently in development and will be available for customers at the end of Q1. For more information please reach out to

No biography entered for this author yet.