AI based diagnosis and support system for cartilage lesion detection on knee MRIs and automated rehabilitation assessment with quantitative biomarkers

Identifying cartilage lesions in patients undergoing MRI of the knee joint has many important implications in routine clinical applications. The MRI is commonly used to assess knee joint, especially cartilage lesions including cartilage softening, fissuring, diffuse thinning due to cartilage degeneration and acute cartilage injury. However, the diagnostic performance depends on the experience of operators and reader, and the different level of expertise also lead to the inter-observer variability in the clinical applications. Moreover, it is very challenging to quantifying these image based biomarkers from MRI due to the variations of their appearances. Therefore, developing a computer based methods for automated detecting cartilage lesions, qualifying and assessing the biomarkers following e.g. knee surgery on MRI would increase the diagnostic accuracy (performance) that would be beneficial to the patients while reducing the inter-observer variability and errors caused by human interpretation. In addition, the designed smart sensors equipped on the patients could provide dimensional information that help assessing the effectiveness of the treatment.

This research proposal is anchored on a recent established collaboration between the Lincoln County Hospital, Prof Lee who is the Director of Research at the ULHT, and the Laboratory of Vision Engineering, School of Computer Science, University of Lincoln. The research aims to develop a fully automated AI based system to detect cartilage lesions and quantify the biomarkers within the knee joint on MRIs. Since the Artificial Intelligence (AI) has shown promising performance in some applications in the industry and has a lot of potentials to apply on a wide variety of applications in medical image analysis. This study will focus on applying the AI techniques on this application to deliver a novel AI diagnosis and treatment (recovery) assessment system of which the efficiency, accuracy and robustness will be validated in the clinical trial.

To achieve this goal, the study is consisted of four phases:
1. Developing the deep learning based anatomic structure (cartilage, bone, muscles) segmentation algorithm. Accurate segmentation is a key step for computer-based surgical planning of interventions affecting the knee.
2. 3D Surface reconstruction /mapping of the segmented knee structures
3. Automatically lesion detection by assessing structural anomalies within the segmented tissues from MRIs for diagnosis and treatment planning.
4. Quantifying the knee joint degeneration using quantitative image-based biomarkers and textual data collected from smart sensors to evaluate the efficiency of treatment.

In these four phases, the accurate segmentation in the phase 1 is a critical task, as it is a prerequisite stage for all the other phases.

The outputs of the phase 2 and 3 would be beneficial to the diagnosis and treatment planning, and could reduce the errors and increase efficiency of the clinical work flow, due to its fully automated nature. Moreover, in the phase 4 we expect to propose a novel biomarker to evaluate the treatment and the stages of recovery based on imaging and wearable smart sensor data.