This repository hosts all code, including exploratory datasets and codebases.
Prosthesis are powerful ways to improve quality of life for an aging population. Most of these need to be operated upon again after 5-10 years or their lifetime. This can be for prosthetic repair, replacement or simply to fix wear and tear.
For this, the surgeon needs to know the correct prosthetic model and manufacturer. This is tough as there are over hundreds of models for each type of prosthetic. For instance, there are atleast 200+ knee prosthetic models which are possible. Identifying this manually, is not only tedious but also error prone.
Identifying the correct prosthetic model make from patient xray
Each prosthetic model make is our target label
Data Constraint: Number of image samples per prosthetic model made is less than 3.
The data constraint effectively rules out most deep learning techniques which require a few thousand images per target label even for fine tuning.
Build an image processing pipeline with two main components:
- Localisation/Segmentation - to extract the prosthetic model outline (countours/edges/polygon matches)
- Feature Extraction - use a feature extraction suite as SIFT, VGG16 or similar
- Classification - in the feature space, find the top 5 most similar models from your database
- Alternatively: Use template matching techniques here to collapse last 2 steps into one
- https://orthoinfo.aaos.org/en/treatment/revision-total-knee-replacement
- http://www.medicalexpo.com/medical-manufacturer/knee-prosthesis-4095.html
- https://www.peerwell.co/blog/2016/10/03/different-types-of-knee-replacement-implants/
The five most commonly used Knee arthroplasty / replacement implants are: PFC Sigma, AGC Biomet, Nexgen, Genesis 2, and Triathlon
- http://www.orthopaediclist.com/category/implants-3.html
- http://www.orthopaediclist.com/category/implant-identification.html
- http://whichorthopaedicimplant.com/
- https://www.realself.com/question/find-type-implant-x-ray-help
Dental Implant:
- Stanford MURA Dataset of Radiographs
- HOG+Linear VSVM with Hard Negative Mining is useful for localization
- Logo Grab Patent
- DSIFT Paper on simaltaneous localisation and classification without training
- An Enhanced Tibia Fracture Detection Tool Using Image Processing and Classification Fusion Techniques in X-Ray Images
- FingerNet: Deep learning-based robust finger joint detection from radiographs
- Development of an analysis system of the X-rays of bones for prosthesis placement
Nirant's Reading List:
- Texture features - GLCM Mean, GLCM Variance, Energy, Entropy, Homogeneity, Gabor Orientation, Markov Random Field (MRF) and Intensity Gradient Direction (IGD)