Vision, Learning and Control Group, University of Southampton

Vision, Learning and Control Group

Medical Image

Medical Image Segmentation

We have been having collaborations with Southampton General Hospital to develop new systems and methods for automated and computerised early diagnosis of lung diseases. In the field of medical image processing, we have therefore been aiming to diagnose lung diseases by processing images taken from lungs using CT-Scans one example of which is shown in the following clip.


The first step of the processing of lungs for an early diagnosis of lung diseases such as COPD (Chronic Obstructive Pulmonary Disease) is to segment them inside the 3D lung CT-Scan images. To do this, we have developed a method known as prior-shape active surface model. The following clip shows some stages of segmentation using this method.



Medical Texture Classification

COPD (Chronic Obstructive Pulmonary Disease) in lungs causes the texture inside the lungs changes. Depending on the severity of the disease, the amount of change in the lung texture varies. Therefore it is possible to find out how much of the lung is affected by the disease and how severe the disease is in the affected areas. Here we aim to capture this change of the texture in lungs by proposing a method based on Gaussian Markov Random Field Models. Our studies show that the histograms of various parameters of this model are good texture descriptors which can be exploited in texture classification. In the following figure, a simple classifier such as K Means Clustering algorithm is used to find the regions of the lung tissues which affected by a kind of COPD known as emphysema.


Medical Feature Detection 

pCLE (Probe-based Confocal Laser Endomicroscopy) is a medical imaging modality to study microstructures within the human organs such as lungs in vivo. Analysis of the images currently requires time consuming manual characterization of regions of interest. Here we aim to create a tool for detection and delineation of alveolar walls and ducts from the pCLE images. In turn this will facilitate automation in vivo measurement of the structural and fluorescence characteristics of the elastin structures opening the door to novel optical biomarkers of diseases affecting areas of the lung that are currently only accessible by surgical biopsy. To this end, we have proposed a method known as parabolic Radon transform. Two examples of the detection of alveolar walls are depicted in the following figure.