AUTOMATIZATION OF COMPUTED TOMOGRAPHY PATHOLOGY DETECTION Semyon Medvedik Elena Kozakevich

Computed Tomography Advanced Diagnostic Tool The scan is performed through taking large amount of 2D X-Rays images. There are two possible way of perform the scan: “Slices” scan “Spiral” scan The images then combined into pseudo 3D image

The Analysis: Manual After CT scan is done, the result images are transferred to expert radiologist. The qualified radiologist examines the scans. All visible pathology found Then analysis is transferred to patient’s physician.

Analysis: Automatic – Why? Analysis is done by human – slow! Analysis is done by human – expensive! Human DO make mistakes! Hence, the analysis is not always detailed

Analysis - Automatic Can we develop the software, that not only perform diagnostic scan, but also will be able to determine the diagnostic by itself?

What we decided to do with that? Develop a method that generalizes all possible CT scans!

HOW? Create averaged image of healthy CT scans Database Average Median Compare input image to the database image Edge detection Distance transform Additional Techniques Segmentation

Results

Brain CT Template Matching Region Oriented Segmentation Edge Oriented Segmentation

Analysis: Automatic Each part of the human anatomy must be observed from different point of view Lungs CT Abdominal CT Liver CT

Lungs CT Gray leveled filters Segmentation precipitation Edge detection Rolling Ball Algorithm

Abdominal CT Active – Contour Approach Center – discover to Region of Interest Size – Use of previous knowledge Average Image Intensity - Each pixel inside found contour effects the results Aspect Ratio - Additional tools to refine the results.

Liver CT Image simplification as preprocessing Detecting a search range – morphological filter Contour based segmentation – labeling based search algorithm

Discussion Each of the described methods uses different approaches Every of the mentioned methods use segmentation as preprocessing. In our experiments segmentation returned the best result. Conclusion: Segmentation is essential for automatic interpretation of the scans.

Why is this issue so difficult? Each of the organs mentioned above has absolutely different structure, hence in the CT scan image it has different shape, intensity and texture. The adjacent organs may have similar attributes as the examined one, therefore the analysis may become very difficult even by radiologist.

Conclusions Medical image analysis continues to be an active area of research. It has many difficult challenges ahead, both in terms of addressing the practical need of community, as well as the theoretical side of CAD.