A CAD System for Screening Xray Chest Radiography

A CAD System for Screening X-ray Chest Radiography Gábor Horváth, Gergely Orbán, Áron Horváth, Gábor Simkó, Béla Pataki, Péter Máday, Sándor Juhász Department of Measurement and Information Systems Budapest University of Technology and Economics and Ákos Horváth Innomed Medical Inc.

OUTLINE § § § The importance of chest screening X-ray or (low dose) CT? The role of computer aided image analysis and detection/diagnostics § The tasks to be solved § Development of a chest X-ray based screening system § § § Goals to be reached Steps of the solution Extensive validation: the first results Present state of the project Further steps MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 2

THE IMPORTANCE OF CHEST SCREENING § In most of the countries of the world there is no X-ray based § § § chest screening The main reason: incidence of TBC is significantly reduced Lung cancer spreads Other diseases of the lung like COPD (chronic obstructive pulmonary disease) are also rapidly spreading There is a need for early detection of these diseases Doubtful rank: according to WHO Hungary is the first in the world concerning the rate of lung cancer among men MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 3

THE IMPORTANCE OF CHEST SCREENING § The goal is of early detection! § Screening for TBC or for lung cancer? § The sensitivity of human (radiologists, pulmonologists) observation (of lung cancer in the early phase) <<100% § Chest X-ray or Low Dose CT? § New digital X-ray systems give much better images (contrast, spatial – 3000 x 3000 pixels – and grey level – 12 -14 bit/pixel – resolution) § Hungary has a whole countrywide lung screening network § Screening for COPD (Chronic obstructive pulmonary ( disease) § (Perhaps integrated with spirometric examination) MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 4

DEMAND FOR NEW DIGITAL SCREENING X -RAY SYSTEMS § X-ray radiation dose is greatly reduced § New possibilities § The possibility of image processing, image enhancement, image manipulation: e. g. extraction of features, noise reduction, intensity adjustment etc, to help medical decision making § The possibility of computer aided analysis, detection of suspicious areas, the application of CAD (Computer Aided Detection) systems § Able to analyse great number of images, without fatigue, without losing attention § § § Remote diagnoses In double reading of X-ray images: second eye Image registration, comparison with archived images MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 5

GOAL OF A CAD SYSTEM n Better nodule detection capability Why is it hard to detect (by computer)? Great variability of images The presence of anatomical noise: shadow of the bones, heart Nodules can be hardly seen in many cases Computers have no anatomical knowledge (and it is rather difficult to teach them) n How to improve nodule detection performance? To improve image quality To filter images (for reducing the noise level) To eliminate anatomical noise To use better nodule detection algorithms MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 7

THE NODULE DETECTION SYSTEM X-ray chest image Mask of the lung area Special filtering CI/AFUM Prefiltering Image features Validation system Gold standard given by pulmonologists Feature selection feature calculation validation classifier image MP-BME World Congress 7 -12 Sept. 2009. Munich result statistics G. Horváth et al. : A CAD System for Chest Radiography 8

LUNG CONTOUR DETECTION The image area to be analysed should be determined Segmentation MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 9

LUNG CONTOUR DETECTION May help to radiologists but surely help to algorithmic processing. Different approaches: The best so far is the ASM based method MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 10

DENOISING - ANATOMICAL NOISE Clavicle and rib cage detection and elimination of the bone shadow Original image MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 11

DENOISING - ANATOMICAL NOISE Clavicle and rib cage detection and elimination of the bone shadow Image without clavicle shadow. Complex approach using active contour models, special filtering and optimization algorithms MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 12

DENOISING - ANATOMICAL NOISE Clavicle and rib cage detection and elimination of the bone shadow Image without the shadow of clavicles and rib cage MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 13

DENOISING - ANATOMICAL NOISE Elimination of the heart shadow Original image MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 14

DENOISING - ANATOMICAL NOISE Elimination of the heart shadow Image without the shadow of the heart. § Modelling the heart § Based on the model the attenuation of the radiation is estimated and compensated MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 15

NODULE DETECTION § Looking for suspicious areas (ROIs) § It may be a rather difficult task MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 16

NODULE DETECTION Looking for suspicious areas Original image

NODULE DETECTION Smoothing and filtering (Convergence index filtering, AFUM) Filtered image MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 18

NODULE DETECTION Image with true and false positive hits

RESULTS ON NODULE DETECTION § Test set: subsection of JSRT database § 124 images with nodules § 20 images where nodule is under the clavicle § Results on candidate selector output § On average 1 less false positive § Maintained sensitivity § Results on classifier output § 15 -18% less false positives while maintaining sensitivity MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 20

RESULTS ON NODULE DETECTION 2 § Results on classifier output § 15 -18% less false positives while maintaining sensitivity MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 21

RESULTS FROC Sensitivity (Free-Response Receiver Operating Characteristic Curve) Number of false positives per image MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 22 22

FURTHER TASKS § Temporal processing registration + subtraction of two-dimensional thorax X-ray images of the same object taken in different times § New algorithms to detect emphysema, COPD (based on lung volume estimation) § Speeding up the algorithms application of graphic processors (GPUs) § Extensive clinical testing MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 23

AN INTEGRATED SYSTEM MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 24

PI - PACS Diagnostics MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 25

The operator’s modul of the whole system MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 26

§ The whole system in Semmelweis University (Budapest) Pulmonological Department MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 27

The flat panel-based X-ray sensor of the whole systems MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 28

Acknowledgement This work was partly sponsored by KMOP-1. 1. 1 -07/1 -2008 -0035 Thank you for your attention MP-BME World Congress 7 -12 Sept. 2009. Munich G. Horváth et al. : A CAD System for Chest Radiography 29