endorsedCompleted

Medication Recognition Machine Learning Pilot

Services Covered:

Custom Software Development

Industry

Information Technology

Duration

1 month

Budget

10K - 49K

Start Date

1 February 2018

Team Size

6-10 Employees

Engagement Model

Nearshore

Project Capability Score (PCS)

68/100

Good

Project Capability Score (PCS) estimates how capable ThirdEye Data is of successfully delivering a project like this, based on its past experience, track record, and ability to handle similar work. Learn more about PCS

Client

Confidential

Canada

Project Summary

ThirdEye Data built a pilot machine-learning system to identify medications from text or photographs for a medical-cannabis related chronic-pain management platform. They implemented a TensorFlow-based approach combined with OCR and fuzzy matching against a 5,000-entry medication list. The pilot clarified the volume of labeled images required for production and delivered components (notably fuzzy matching) that were integrated into the client's product.

Key Challenges

Determine medication and dosage from user-entered text or a smartphone photo
Reduce manual typing by automatically recognizing medications from photos or approximate text input
Validate whether a machine-learning approach could serve as the basis for future platform features

Project Deliverables

Machine learning model for medication recognition
OCR integration for label extraction
Fuzzy matching system against a 5,000-medication database
Augmented training dataset (image multiplication) for model training

Project Solution

ThirdEye Data proposed and implemented a TensorFlow-based machine learning approach combined with optical character recognition (OCR) libraries and fuzzy-matching logic. They trained models by augmenting a public pill image dataset and built backend processes to accept uploaded photos, extract text via OCR, and fuzzy-match results against a 5,000-medication list to provide informed guesses for medication and dosage. The engagement delivered deployable components and documentation enabling the client to integrate parts of the solution into their product.

Project Outcome

Realized large data requirements needed to productionize medication-recognition ML models
Augmented a public pill image dataset by multiplying images for training
Deployed parts of the technology (especially the fuzzy-matching component) into the client's product
Strong communication, documentation, and rapid delivery facilitated integration of deliverables

Platforms

AI/ML Platform

Tech Stack

TensorFlow

Client Endorsement

Overall Review Rating

4.75

Timeliness

Cost Rating

Willing to Refer

Quality of Deliverables

“This pilot project served for recognizing the amounts of data required for future machine learning ventures, though several parts of the tech they developed were deployed in the current software offering. Easy communication, thorough documentation, and speed of delivery are highlights of their work.”
Anonymous
VP of Technology
Date: 03/06/2019