AI DRIVEN LUNG WORKFLOW
Company: Echonous is a medical device startup for miniature ultrasound devices.
​
Context: Traditionally, medical practitioners use stethoscope and chest x rays to assess lung health. Chest X rays suffer from generally low sensitivity and specificity as well as a lack of immediate availability of results.
​
Dr Daniel A. Lichtenstein and Dr Michael Blaivas are pioneers and have published multiple papers in how ultrasound can be used to detect common lung conditions (pulmonary edema, pulmonary embolism, pneumonia, and pneumothorax) using BLUE protocol at the point of care.
​
We aim to enable primary care physicians to perform these lung assessments without having to send the patient to the Xray labs.
​
Project goal: Create a lung workflow derived from BLUE protocol for clinicians which aids in the diagnosis of diseases such as pulmonary edema, pulmonary embolism, pneumonia, and pneumothorax.
​
What is BLUE Protocol?
Bedside lung ultrasound in emergency (BLUE) is a protocol for the immediate diagnosis of acute respiratory failure with diagnostic accuracy of 90.5%. (Add Source ).
​
​
​
STEPS FOR AI DRIVEN LUNG WORKFLOW
Worked closely with Dr Blaivas to come up with this workflow.
STEP 1
Select Num of Zones
Each lung scan is divided into multiple zones. There is no standard on a number of zones. A larger number is going to be more thorough but time-consuming.
STEP 2
Record a clip for each zone. Analyze in real time for A-lines, B -lines and lung sliding if possible.
Record Clip / Analyze
STEP 3
Post processing of recorded clips
System looks at all the collected clips and performs ML and AI algorithms to aid in diagnosis to the clinician.
STEP 4
Show the artifacts and results. Allow clinician to see all the clips and make an informed diagnosis.
Show Decision Tree
Low fidelity prototype
This screen shows all the partitions of the lung and various artifacts found in each partition.
Prototype Video
I created this lung workflow that allows the clinician to customize their exam based on their personal protocol.
The images go through AI and ML algorithms and clinician can make more informed decisions based on the diagnostic recommendation.
​
We are using this video to gather feedback from the doctors.
​
Voice credit - Simon Kim
WHAT'S NEXT
Currently in the discovery stage, I am working with AI engineers to investigate and develop the AI models needed for the automation.
I created this data collection protocol for the sonographers to collect images of the lung in various positions of the patient.
​
The images collected from this protocol is being used by AI engineers to train the model.
​
Based on the learnings from the dev team, the plan is to revisit the protocol, and plan the phase 1 of development activities.
