Computer Vision News - July 2021

2 Summary 16 Medical Imaging Application Pulmonary Embolism (PE ) is a life-threatening condition with a mortality rate of up-to 30%, where an embolus blocks one of the pulmonary arteries. When the thrombus originates in other blood vessels (usually deep vein thrombosis) it is defined as acute PE. Occasionally the clot develops over time and inseparable from the vessel wall, slowly increasing the artery blockage. This is defined as chronic Pulmonary Embolism . P E treatment needs to be immediate and accurate, especially in acute PE. Common diagnostical procedure requires CT Pulmonary Angiography (CTPA) , due to its availability, speed, and accuracy. A high-resolution 3D scan of the pulmonary arteries assists in PE detection. However, precise detection of the emboli can be challenging – it is difficult to track pulmonary arteries throughout the scan, and errors are often made. RSIP Vision has developed an AI-based Airways Segmentation suite. This system robustly segments lungs, lobes, nodules, airways, and pulmonary blood vessels from CT scans. Pulmonary arteries and airways run alongside each other in hierarchical bifurcations from lobes to secondary pulmonary lobule, suggesting that knowledge of airways structure can assist in blood vessel segmentation and emboli detection. The above introduction emphasizes the need for an automatic PE detection tool . Combining Artificial Intelligence and classic computer vision algorithms can efficiently and accurately detect the location of the emboli based on the CTPA scan. This system will maximize the segmentation accuracy of the pulmonary arteries as they will stand out compared to the rest of the pulmonary structures, making emboli detection easier. The main challenge of segmentation from CTPA is overcoming image artifacts originating from movement (breathing), implants, human error, and anatomical differences. Advanced neural networks have proven capabilities of achieving a trustworthy solution despite poor image quality. Apart from detecting the vessel filling defect, additional clinical data is necessary for treatment selection. Enlargement of the right atrium or ventricle, as well as dilation of the pulmonary arteries influence the physician’s decision-making. Segmentation of the cardiac chambers and proximal pulmonary arteries is feasible from the CTPA using similar deep learning techniques , and automatic detection of abnormalities can be calculated and provided as an input for clinical decision-making. Pulmonary Embolism Detection Using AI