awicklund / endoobjdetect

A TensorFlow-centered approach to object detection using Faster R-CNN to detect endometriosis on the GLENDA database

EndoObjDetect

This repository is HIPAA compliant. All sensitive and identifying patient information has been removed prior to publishing.

A TensorFlow-centered approach to object detection using Faster R-CNN to detect endometriosis on the GLENDA database

Model Version: 2.0

Model used: Faster R-CNN Inception ResNet V2 1024 x1024

Implemented on Google Colab on High-RAM, GPU-accelerated environment

Some of the images that were produced:

Endometriosis is a painful condition where the tissues produced inside the uterus to protect a fertilized egg grow outside the uterus. This is the first publically available image study of endometriosis and the model is designed to detect endometrial tissues on and around the uterus.

The loss graphs are provided by Tensorboard as the model finishes training:

The files are organized as follows:

Finished examples - contains 10 example images and 10 doctor-annotated example images (alpha channel)

Graphs - graphs displaying model information provided by Tensorboard

Model - contains both important outputs (Endo), as well as test data (Endo) and inputs (my_tfod)

Endometriosis Modeling 101 - user-friendly document that describes the model for the general public

My_Code.ipynb - the Jupyter Notebook that I used to create the model

Slide_Deck.pptx - the Powerpoint used in presenting the model

Technical_Documentation - a write-up designed for data scientists, contains technical jargon

Tutorial_Used - the tutorial I referenced in creating and running the model

Got a question? Or a neat project to share? E-mail me at [email protected]