Visual Computing Techniques for Automated Detection of Osteoporosis and Osteoarthritis. The project OSTEON is centered around analyses of high-resolution radiographs of human bones (currently calcaneus, knee) with the focus on automated detection of osteoporosis and osteoarthritis.

Osteoarthritis (OA) is a group of mechanical abnormalities involving degradation of joints, including articular cartilage and subchondral bone. In order to identify OA cases, physicians need to visually inspect a number of radiographs every day. Moreover, looking at the projected bone texture even trained expert eyes are often incapable of making accurate diagnoses or early predictions. To reduce the human labor, automated detection should help screening for obvious OA cases and flagging suspicious cases for a closer examination. The challenges being solved within the scope of this project cab broadly be divided in medical image processing tasks and tasks related to machine learning.

The goal of the medical image processing part is to automatically locate and label landmarks, contours, and regions of interest, consistently across large number of input radiographs. The automatically found shapes are transformed into attributes, such as distances, areas, and angles. The regions of interest are used to measure texture properties of the underlying bone. Typical examples include fractal measurements, frequency analyses, or texture descriptors commonly used in computed vision.The aim of the machine learning part is to map the automatically extracted attributes to the human expert scorings. These scores may include two classes, e.g., healthy versus OA, several classes, e.g. scores proposed by Kellgren and Lawrence (grades of Osteophytes, progress of bone sclerosis) or a range of continuous values. The aim is to find suitable models (i.e., classifiers) that best reflect the expert scoring.