Computer-assisted diagnosis of early-stage pleural mesothelioma using 3D data - system development and validation

Status:

completed 03/2013

Aims:

The early detection of pleural mesothelioma for asbestos-exposed individuals requires regular checkups. Pleural thickening can be an indicator for this disease and detected in thoracic 3D CT data. A follow-up assessment is performed by measuring the thickening growth and allows an estimation of development in case of possible pleural mesothelioma. A manual analysis of the image data is time-consuming and its results are subject to inter- and intra-reader variability. To deal with the increasing number of pleural mesothelioma cases, a computer-based system for the automatic detection of pleural thickenings was developed. The system detects changes of the thickenings quantitatively and reproducibly documents these findings.

Activities/Methods:

Special methods were designed for the precise thickening detection such as automated thresholding, probability-models, morphological analysis and multiscale analysis. After an optional manual check, the detected thickenings of two points in time are compared. For the full-automated processing, a 3D surface-based registration technique is applied. The lung images are superimposed minimizing the image error in terms of the segmented lung tissue. This allows to locate the thickenings and to match them between two points in time. The extracted relative position of the thickenings and information about their tissue can be used to match the thickenings. A user interface has been implemented, which enables the diagnosis, the correction, and the documentation of the data from two points in time. The evaluation was performed in 3 stages. A total of 20 CT data sets from 10 patients were considered.

Results:

Firstly, the evaluation of the detections showed that 73% of detected swellings were correctly detected. Secondly, the fully-automatic assignment of the thickenings is correct in 95% of cases.

At the end, the evaluation of the final system quantitatively confirmed that progression of the disease can be reliably determined with the help of the software system. A comparison to fully manual diagnosis showed that progression, which was recognized by at least two physicians, could not be recognized by the software for two cases. However in three cases, progression, which was recognized by only one or no reader in manually diagnosis, was signaled through the system. Thus, using full-automatic diagnosis and correcting the results by an expert reduces inter-reader variability.

For the first time, a fully automated software system allows the physician to quantitatively and reproducibly measure the change of pleural thickenings as well as its documentation. The software thus provides an added value with respect to reproducibility and sensitivity in the detection of progression. Through the intuitive user interface and automation, the required time for diagnosis will reduce in practice. The developed system provides not only a useful tool for early detection of the disease in the screening process, but it helps to quantitatively document the diagnosis and the progress of the disease, in cases of a pre-existing disease.

Last Update: