Detection of emphysema in patients with COPD.

Pino Peña, Isabel

4. term

Biomedical Engineering and Informatics, Master

2013

2013-06-04

48

Kronisk obstruktiv lungesygdom (KOL) er en langvarig lungesygdom, der begrænser luftstrømmen. Den omfatter kronisk bronkitis og emfysem, hvor lungeblærerne er beskadiget. I dag vurderer radiologer ofte emfysemets sværhedsgrad ved at give visuelle scorer af CT-scanninger af brystet, hvilket kan variere mellem fagfolk. Dette projekt undersøger en objektiv, automatisk metode til at opdage og kvantificere emfysem ved hjælp af teksturanalyse af CT-billeder. Lungerne blev først segmenteret med en region-growing metode (en teknik, der grupperer nabopixels i lungeregioner). Derefter blev teksturtræk beregnet med en gråtoners medforekomstmatrix (co-occurrence matrix), som beskriver, hvor ofte forskellige nuancer optræder sammen og dermed fanger billedmønstre. Disse træk blev brugt til at træne en support vector machine (SVM), en maskinlæringsklassifikator, til at skelne mellem normalt lungevæv og emfysematøst væv. Metoden blev trænet og evalueret med leave-one-out-krydsvalidering på scanninger fra 11 deltagere (9 ikke-raske og 2 raske). For at validere resultaterne sammenlignede vi de automatiske scorer med visuelle scorer fra to radiologer og én læge. Overensstemmelsen blev målt med kvadratisk vægtet kappa og var rimelig. Resultaterne tyder på, at metoden kan kvantificere emfysemets sværhedsgrad og kan hjælpe med at reducere forskelle mellem vurderinger. Den er et lovende bud på automatisk kvantificering af emfysemforandringer hos patienter med KOL.

Chronic Obstructive Pulmonary Disease (COPD) is a long-term lung condition that limits airflow. It includes chronic bronchitis and emphysema, where the air sacs in the lungs are damaged. Today, radiologists usually judge emphysema severity by visually scoring chest CT scans, which can vary from one expert to another. This thesis explores an objective, automatic way to detect and quantify emphysema using texture analysis of CT images. The lungs were first segmented with a region growing method (a technique that groups neighboring pixels into lung regions). We then computed texture features using a gray-level co-occurrence matrix (a summary of how often different shades appear together, which captures image patterns). These features were used to train a support vector machine (SVM), a machine-learning classifier, to distinguish normal lung tissue from emphysematous tissue. The method was trained and evaluated with leave-one-out cross-validation on scans from 11 participants (9 non-healthy and 2 healthy). To assess validity, we compared the automated scores with visual scores from two radiologists and one physician. Agreement was measured with quadratic weighted kappa and was found to be fair. These results indicate that the proposed method can quantify emphysema severity and may help reduce differences between readers. It is a promising approach for automatic quantification of emphysema lesions in patients with COPD.

[This abstract was generated with the help of AI]

Image Analysis ; Texture-based ; Region Growing ; Emphysema ; COPD

Download
View record in AAU Student Projects