Knee Shape Analysis

Jessie Thomson


Osteoarthritis is a degenerative disease that effects the entire joint, degrading articular cartilage and deforming the surrounding bones and tissue of the affected joint. The disease affected 8.5 million people in the UK in 2012, and caused over 2.6 billion annually in analysing and repairing the joint, and alleviating the pain caused. Osteoarthritis is a severely debilitating disease across the population. With no known pathogenesis, research emphasis lies with finding new and improved ways of supplementing the effects of the disease.

This study looks at the area of the knee, which is one of the main areas afflicted by Osteoarthritis. The project aims to analyse the mechanics of the disease in the joint to identify significant markers for Osteoarthritis and related knee pain. These markers will be based on various shape and texture features extracted across the whole of the joint. Current clinical methods rely on manually applied, semi-quantitative descriptions of the osteoarthritic features to detail the severity of the disease. Due to the semi-quantitative nature of these methods and the weighting of multiple OA features per grade, this can often be at risk of subjective beliefs of the grader. Leading to discerpencies and reliability issues when clinicans are analysing large sets of radiographics images. The application of automated methods helps to solve this issue with subjectivity, through the use of standardised, quantitative, rules and measurements. Due to the multifactorial nature of Osteoarthritis, the project looks at multiple explicit and implicit features relater to OA. These features include: overall shape, osteophytes, trabecular structure, tibial spines and inter-condylar notch, and joint space.

The project uses a 74 point model manually annotated on 500 images (37 points each for both the Tibia and Femur) and an RFCLM algorithm to find these points in a new image.

74 Point Knee Model

Shape Features

The shape features extracted from the knee are taken from a series of annotated points along ythe respective feature. The annotated points are then used to train a Constrained Local Model (CLM) algorithm which picks up on the main modes of shape variation across the training set. The moving images in the sections below show the 3 highest patterns of variation.

Overall Shape

The overall shape features are taken from the 74 point model (above).


Taken from 48 points outlining the joint margins (medial tibia, medial femur, lateral tibia, and lateral femur).

Joint Space

A joint space shape model (JS-SSM) is built from 40 points (10 per joint space surface).

Texture Features

Image gradient information is taken from sections of the radiographs. These regions are placed using the 74 points (above) found in the image.


Trabecular Structure

Tibial Spines and Inter-condylar Notch


The features were used to classify OAI radiographs in 4 experiments: current OA, current pain, later onset OA (from baseline images), and later onset pain (from baseline images). The results show that current OA achieves the highest AUC and accuracy, with some correlation found between radiographic features and pain/later onset outcomes.
Experiment AUC (95% CI)
Current OA 0.9035 (0.897 - 0.91)
Current pain 0.6629 (0.65 - 0.675)
Later onset OA 0.6137 (0.587 - 0.64)
Later onset pain 0.6085 (0.594 - 0.623)


J. Thomson, T. ONeill, D. Felson and T. F. Cootes, Automated shape and texture analysis for detection of Osteoarthritis from radiographs of the knee, Proc. MICCAI 2015, Part 2, pp.127-134

J. Thomson, M. Parkes, D. Felson, T. ONeill and T. F. Cootes, Automated multi-feature analysis of current and future onset pain in osteoarthritic knees, Int. Workshop on Osteoarthritis Imaging, (abstract).

J. Thomson, T. ONeill, D. Felson and T. F. Cootes, Detecting osteophytes in radiographs of the knee to diagnose Osteoarthritis Proc. MICCAI MLMI 2016 (to appear).

Under review: L. Minciullo, J. Thomson, T. F. Cootes, Combination of Lateral and PA View radiographs to Study Development of Knee OA and Associated Pain, SPIE 2016

In preparation: J. Thomson, M. Parkes, D. Felson, T. ONeill and T. F. Cootes, Automated Radiographic Multi-feature Analysis of Current and Future Onset OA and Pain, Osteoarthritis and Cartilage