A hybrid lung nodule detection scheme on chest x-ray images

Orbán, G [Orbán, Gergely Gyula (információs rends...), author] Department of Measurement and Information Systems (BUTE / FEEI); Horváth, G [Horváth, Gábor (Intelligens rends...), author] Department of Measurement and Information Systems (BUTE / FEEI)

English Conference paper (Chapter in Book) Scientific
    Identifiers
    In the current study, we propose a computer aided detection (CADe) scheme targeting lung nodules on chest radiographs. Instead of using the common scheme of a nodule enhancement filter followed by a classifier, our novel approach utilizes separate filters targeting different types of lung nodules. Smaller and low-contrast nodules are enhanced based on the convergence of gradient vectors, while larger objects with higher contrast are to be found by a modified top-hat filter on a local contrast enhanced image. After segmentation of nodule candidates and calculation of features on them, a classifier network consisting of three Support Vector Machines (SVM) reduces the number of false positive findings and merges the results of the two enhancer filters. The CADe system is tested on a radiograph database containing images of 93 patients with validated lung nodules and 150 healthy cases. The results of a system using only one nodule enhancer filter and the hybrid system are compared using a Free-response Receiver Operating Characteristic (FROC) analysis. The hybrid solution turned out to be clearly superior to the other schemes. Considering the whole system, we experienced 72% sensitivity at a false-positive rate of 2 and 77% sensitivity at a false positive rate of 3. © 2011 Springer-Verlag Berlin Heidelberg.
    Citation styles: IEEEACMAPAChicagoHarvardCSLCopyPrint
    2025-04-26 07:43