Fully automated system for the analysis of the efficacy of knee replacement surgery

Musculoskeletal (MSK) diseases affect about 16% of adults and more than 30% in the age group 65+. As the incidence of most MSK diseases increases with age, the socio-economic impact of MSK diseases is increasing steadily in countries with an ageing population such as the UK. During 2014-2016, more than 260,000 primary knee replacement surgery (KRS) procedures were performed in England, Wales and Northern Ireland. KRS may, in the short or long run, lead to complications that will require a revision KRS and may result in life-threatening conditions (e.g., caused by a loose implant). Currently, the risk of KRS failure is 11% ten years post-operatively. When a KRS fails it is important to schedule a revision KRS to remove a loose or damaged implant before irreversible harm is done to the knee joint.

There are an increasing number of medical images being gathered in all UK NHS hospitals, with the growing need and opportunity to utilise this information to improve the health of the nation. Currently, medical images are underutilised in clinical practise and research into MSK diseases where 2D radiographs are the imaging technique of choice due to wide availability, speed of acquisition and low cost.

The aim of this project is to develop an automated software system to study the effectiveness of KRS. The system will automatically locate the knee joint in radiographs and analyse the radiographic shape and appearance of the knee bones or implant. It will then combine the obtained image-based data with clinical data aiming to predict and identify signs of early joint failure. The goal is to transform clinically collected image data into useful medical information to benefit healthcare at individual and societal levels.

Various clinical placements will be undertaken to identify clinical needs and analyse the clinical workflow in KRS decision-making. This involves finding answers to questions such as how can the current workflows be revised to fit a new software system. Working closely with the Connected Health Cities project will provide details on how to technically implement such a system into the existing digital healthcare infrastructure. The collected information will be used to inform a suitable design (e.g., graphical user interface) and implementation strategy for the system to be integrated in the clinical setting. The software system will be validated at various stages of the development cycle to ensure that it fits the purpose. The usability and acceptability of the system will be evaluated during a pilot trial towards the end of the project.

This research will result in a computer-aided system to inform the KRS decision-making process and the resulting medical management, improving the quality of care in clinical practice.

In the long-term, this project aims to improve the health of the general population through a greater understanding of early joint failure in KRS, and by providing a computer-aided system for identifying and monitoring the latter in clinical practice. This work will have a positive impact on: (i) patients, who will benefit from better treatment choice and management, leading to improved quality of care; (ii) clinicians, who will have access to additional data to make an informed decision and who will save time by using an automated system to generate such data; (iii) the healthcare system, which will benefit from cost reductions; and (iv) the wider research community, where newly developed imaging and data modelling methods will contribute to advancements in other areas of automated image analysis as well as large scale data analysis.

This project aims to develop a computer-aided system to analyse radiographic signs of early joint failure in knee replacement surgery (KRS).

The software system to be developed will be based on the BoneFinderTM-technology (www.bone-finder.com) which uses machine-learning methods to automatically locate skeletal structures in radiographs. Based on a set of training data consisting of radiographs and feature point annotations, BoneFinderTM uses Random Forest-based shape model matching methods to locate the corresponding feature points in unseen images. BoneFinderTM has been applied to automatically outline knee joints in radiographs but further methodological development is required to account for disease and to incorporate implant detection.

For a given image, the automatically identified feature points will be used to quantify the knee bones and implant fitting. This will include conventional geometric measurements of lengths and angles but these only give a sparse representation of the structures. Statistical Shape and Appearance Models will be used to capture and quantify their overall shape and texture. Using retrospective data, machine-learning methods will be applied to combine the derived radiographic shape and appearance information of the knee joint with collected clinical data to analyse signs of early joint failure. This will lead to a computer-aided system that will have learned to identify radiographic indications of early joint failure based on radiographs and clinical data for any new subject, informing the KRS decision-making process.

A feasibility study will be conducted to identify the clinical requirements for usability and acceptability, and for the integration of the system into the clinical workflow. This will also include the identification of a suitable implementation strategy of the system into the existing digital healthcare infrastructure. The system will be validated during a pilot trial.