List of questions

One of the advantages of MRI is the ability to map regional changes in physiological parameters throughout the organ of interest. The commonly used approach based on global average across the organ of interest may lead to information loss from small regions in which biological relevant change occurs, or there could be compensatory changes in other regions that counteract the changes in the region of interest. A robust method which addresses statistically significant heterogeneity within the sample is of great importance to detect relevant changes in pharmacological interventions.

If one region of the lung is damaged by, for example, smoking, the loss of function may be compensated by other regions of the lung. This reduces the sensitivity of global lung function measurements particularly in early disease. Medical imaging techniques can detect these regional changes but there is a need for a predefined endpoint for use in clinical trials. There are a large number of potential endpoints to described a spatially heterogeneous response.

Active Shape Models can be used to segment 3D structures such as muscle, lung and brain in medical images. This technique does not appear to be widely available despite its obvious appeal. Is there a user friendly way to access this technology to apply to a wide variety of segmentation problems?

Hur definierar och mäter man homogenitet på hyperspektrala bilder?

What kind of concept-based features (descriptors) can be extracted in a real-time video for non-rigid objects tracking/ recognition? The extracted features must adapt to large variance in illumination, pose, scaling, and view angle.

How to detect/classify abnormal human behavior in a real-time video surveillance system?

Which applications in a video surveillance system urges for depth information?

How will advances in CT/MRI image analysis & segmentation change the way we treat patients in the future?

Is "ensemble CAD" a future path for medical image analysis?

Despite the progress in image analysis research, Computer-Aided Detection/Diagnosis still struggle to find its way into clinical practice in many areas. Where does research stand today when it comes to combining several sub-optimal methods, into an ensemble method that can reach levels of sensitivity and specificity that no single method can achieve? This would be similar to how multiple prognostic methods are always employed in meteorology.

Is Computer-Aided Simple Triage (CAST) a sweet-spot for medical image analysis?

A challenge for any Computer-Aided Detection (CAD) system is that the false positives generated entail additional clinical work that counteracts the wanted efficiency benefits. CAST represents a different way to deploy automatic analysis, one possibility is as a pre-processing step triaging away normals. As long as sensitivity is 100%, i.e. no false negatives, this could reduce the inflow to more costly diagnostic work, even with far from optimal specificity. CAST is now mentioned mostly in emergency care, but should CAD research in diagnostic imaging also further embrace this type of usage?