USC Healthcare cut down processing time for clinical trials by 75% using Machine Learning and Google Cloud Vision

In the spirit of continuous improvement, we wanted to keep the momentum going. We knew we had an opportunity to improve further. So we turned to Google Cloud and machine learning. Pluto7 helped us see the potential of MLAI by working on one piece of the process. In a proof of concept project, we applied the technology to the part of the budgeting process called Medicare Coverage Analysis.

– Allison Orechwa, Director of Programmatic Development, University of Southern California


University of Southern California is a large private university and academic medical center in Los Angeles, CA. They collectively conduct over 200 clinical trials each year. They have strong research programs in cancer, neuroscience, biomedical imaging, stem cells and more. Goals of research are to better understand diseases and create new therapies and standards of care for patients and the public at large. The Clinical and Translational Science Institute is charged with improving the research pipeline at USC to get more treatments to more patients more quickly. Given the local population, USC brings expertise in diverse populations to the national network of institutes. If research is done with only one segment of the population, USC won’t know how well the new treatment will work for the rest of the population. They aim to correct that by getting more LA citizens involved in research and making the research faster and more effective.

Why We Chose Pluto7

Pluto7, a Google Cloud Premier Partner, specializes in deploying accelerated solutions and custom applications in smart analytics, machine learning, and AI. The client realized that we can add immense value by leveraging our knowledge of machine learning and integrating that with current Google Cloud Platform tools such as BigQuery, Natural Language, and Google Kubernetes Engine to create an integrated solution that allows them to leverage their industry experience to build a compelling, data-driven product for many SMBs and middle market customers.


Two years ago USC surveyed it’s researchers. Many people complained about administrative burdens like too many systems, unclear requirements, long activation times. Within the long lifecycle of a clinical trial, there is a long pause in the actual science for approvals and logistics. For example, ethics committee, setting up pharmacy orders, negotiating contracts with pharmaceutical company that’s paying for the study. This “activation” process takes 10-20 people over 80 man hours to complete with lots of back and forth and wait times, which comes out to 3 months on average.

They knew they had an opportunity to improve further. So USC turned to Google Cloud and machine learning. Pluto7 helped them see the potential of ML/AI by working on one piece of the process. In a proof of concept project, we applied the technology to the part of the budgeting process called Medicare Coverage Analysis. 

Pluto7 started by creating process workflows and uploading Data to begin exploratory analysis. After that we analyzed data formats, evaluated conversion methods, Used OCR for extraction, create document pipeline. Once the preparation was finalized we Implemented NLP for pattern extraction, trained & tested 3 ML Models (Naives Bayes, KNN & SVM). After 2 iterations and validations Pluto7 finalized QCT automation & evaluation, and captured limitations & discrepancies.


At Pluto7, we have provided a solution demonstrating to the Keck School of Medicine of USC how ML/AI can be leveraged to build a Medicare Coverage Analysis (MCA) in less time than a human-only process. Over the last few years USC has achieved phenomenal success in transforming the speed and efficiency in the way clinical trials are processed. USC has reduced the total clinical trial average cycle time, one of the best in the industry today.

Industry Healthcare| Higher-edu

Solutions Education-ml| Healthcare-ml

Organization Name: Keck School of Medicine of USC


  • Extracting subscript or superscript text from calendar headers or procedures
  • Extracting footers as part of calendar output
  • Questions on Primary objective extraction needed additional refinement
  • Requirement of a larger dataset and analysis- capturing all procedures, detecting & correcting outliers are ways to improve accuracy and reduce errors.


  • Reduced time to complete by 75% for QCT and Calendar protocols
  • Labeled billing designation with 90% accuracy
  • Predicted procedure required with 75% accuracy

Products Used

  • Google Cloud Platform 
  • Google Cloud Storage
  • Cloud Vision API
  • Cloud Machine Learning Engine

Customer Success Stories

Talk to an Expert

Transform your business by leveraging the power of Machine Learning Artificial Intelligence, Analytics, and IoT solutions.

Contact Us