S54 Automating the analysis of thoracic ct scans in cystic lung disease

IntroductionCertain disorders of the lung, such as Birt-Hogg-Dubé syndrome (BHD) and lymphangioleiomyomatosis (LAM), are characterised by the presence of multiple pulmonary cysts. Radiological analysis using thoracic computed tomography (CT) is the mainstay of diagnosis and follow-up of these disorders. The rate of change of the cysts contributes to therapeutic decisions including the prescription of potentially toxic therapies, most notably mTOR inhibitors in LAM. At present, cyst parameters including their location, size, shape and number are determined by the review of CT images by radiologists. Despite expert training, this process is prone to human error and susceptible to inter-observer disparity.ObjectiveWe wished to determine if automation of cyst analysis could provide robust data to aid the radiologists in their reporting of thoracic CT scans.Methods and ResultsSoftware was developed using C++to extract data from standard Digital Imaging and Communications in Medicine (DICOM) CT files. For each scan, voxels in lung parenchyma or cysts were detected by radiodensity being in the range from −935 HU to −610 HU or below −935 HU, respectively. The 3D-cyst boundaries were identified by means of novel recursive algorithm (figure 1). Trachea and airways were automatically detected and excluded from further analysis. Number of cysts per patient was recorded and each cyst analysed in terms of volume, spatial location, sphericity and cylindricity (calculated by using eigenvalues of gyration tensor for corresponding cyst) and opacity. The software was calibrated empirically through iterative adjustment of the above threshold values and comparison with scores generated by an expert thoracic radiologist thus enabling the reliable differentiation of cysts from noise. As proof-of-principle, the scans of 10 individuals with BHD and 10 with LAM were analysed in a blinded manner by the computer and compared with independent radiology reports.ConclusionAutomated image analysis provides a new set of objective cyst parameters and offers added value to the thoracic radiology reporting process. Future studies will determine the relative sensitivities of human vs. automated CT analysis in the diagnosis and monitoring of cystic lung diseases including BHD and LAM.Abstract S54 Figure 1Automated detection of lung cysts. 3D rendering of the thoracic CT scan of an individual with BHD following automated detection of cysts but prior to deletion of airways. Note irregular cysts