Case Studies

Artificial Intelligence & Deep Learning

Theta Tech was tasked with creating a brain navigation software for a medical AI lab. We were responsible for taking baseline AI algorithms that detects brain cancer and enhance the system to meet production standards that would be used by neurosurgeons.

‍

  • Ability to handle various medical image data types
  • Interfaced with existing hospital software
  • Employed modern deep learning and analysis frameworks
  • Determined best position for AI technology and associated clinical trials for FDA approval
  • Ensured ease of use by clinician

AI Applications

  • Designed algorithm to handle various data types
  • Scaled up lab level software to production requirements

Data Analyzed

  • 3D brain MRI results
  • 3D volume meshes of tumors with either tissue necrosis or cancer recurrence

Technology Used

  • Python
  • Docker images
  • Deep learning and image analysis frameworks

Strategy & Execution

  • Communicate with medical AI lab on requirements
  • Visualizing clinical results in 3D
  • Working with existing engineers and algorithms
  • Setting up cross-validation experiments
  • Reporting accuracies, sensitivity, specificity, Type 1 & 2 error
  • Design system around existing hospital systems and with end user in mind
Theta Tech developed a 3D volume overlay on an MRI scan.
Colors represent the probability of cancer development.

Biostatistics

Theta Tech has performed ongoing data analytics for a large, U.S. based prostate cancer clinical trial. We worked directly with researchers and clinicians to fine tune the automatic report that our system generated.

‍

  • Standardized various unique data formats and medical data streams
  • Analyzed for correlations between multidimensional data
  • Utilized in house statistical analysis for predicting cancer recurrence and treatment efficacy
  • Data was presented using tables and graphs

Data Analyzed

  • Protected health information
  • Patient bloodwork
  • MRI results
  • Biopsy results

Technology Used

  • Matlab
  • Excel statistical package

Strategy & Execution

  • Received anonymized HIPPA data
  • Identified new correlations between orthogonal data streams and clinical outcome
  • Processed clinical trial data and worked closely with researchers and clinicians
  • Agile turnaround with final reports for stakeholders
Prostate specific antigen (PSA, ng/mL) analysis completed by Theta Tech.

Image Analysis

Theta Tech was tasked to create an image analysis system capable of processing 100,000 - 500,000 images per month. The end result was a system that allowed the user to run experiments to tune the parameters live and further enhance its accuracy.

‍

  • Built custom AI that sent results to the client
  • Used AWS to host convolution neural network deep learning system
  • Implemented feedback loop to auto-retrain the AI system
  • Full stack solution - WebApp frontend connected to C++ deep learning backend
  • Seemless integration with client's existing systems
  • Full user control over system with reported accuracy values (FP's, FN's, PPV, NPV, Total Accuracy)
  • Scalable and required minimal maintenance

AI Applications

  • Utilized custom convolution neural network deep learning system
  • Custom "Mixture of Experts" neural architecture
  • Created robust AI system that was generalizable to unseen real world data
  • AI was automatically retrained when new, human-labeled data was fed into the pipeline
  • Bridged together the AI system with the frontend interface

Data Analyzed

  • Images from a proprietary company database
  • Images scraped from the internet
  • Human-labeled training data

Technology Used

  • AWS (EC2, SQS, S3, Load Balancer)
  • C++ with Caffe deep learning framework
  • Python (Amazon Library)
  • Spring framework written with Java for backend
  • JavaScript for web portal frontend
  • Custom MySQL schemas
  • JSON data for communication

Strategy & Execution

  • Worked directly with CEO to create a custom algorithm for the business
  • Worked with CTO to integrate AI system into existing tech stack
  • Optimized AI system to minimize server cost
  • Auto generated reports on accuracy and efficiency
Image analysis algorithm that Theta Tech developed.
The "Lee" text on the top left is searched within the given image.

Medical Devices

Theta Tech has worked with medical devices in varying stages of development and product development experts on software and hardware projects.

  • Regularly interfaced with regulatory experts for engineering projects
  • Upgraded heart abnormality detection algorithm for a heart rate monitor
  • Created a battery operated, inexpensive, handheld confocal microscopy device for detecting skin cancer and esophageal cancer

Regulatory Correspondence

  • Close communication with FDA consultants
  • Researched medical predicate devices
  • Communicated prototype deliverable with an academic technology transfer office

Engineering

  • Iterated through several fluorescent confocal microscopy prototypes using 3D printing
  • Processed medical signal data and linked it to client's existing device code
  • Created algorithms that analyzed EEG signals
  • Installed custom algorithm and ported into C++ to work with heart rate monitor devices

Strategy & Execution

  • Worked with CTO on code and product requirements
  • Developed highly effective and low cost final device