No abstract

SUMMARYThis paper proposes an automatic method for extracting liver tumors from three-dimensional abdominal CT images of four phases (noncontrast, early phase, portal phase, late phase). In the proposed method, the liver region is extracted from the image of each phase. Enhancement process is applied to the region by using a three-dimensional adaptive convergence index filter. The local maximum points are extracted from the enhanced image and the candidate regions for the liver tumor are extracted by applying the region-growing process. Multiple features are derived from the extracted candidates and it is decided whether or not the candidate actually corresponds to a tumor, based on the Mahalanobis distance calculated from the features. The processing from the enhancement to the discrimination of the candidates is performed independently for each phase, and four processed images are generated. Finally, the resultant images are combined and output. The proposed method is applied to 40 three-dimensional CT images of 10 patients taken by a multislice CT scanner. The result is promising, since all 8 existing tumors were detected, and the number of false-positive regions per image was 0.

No abstract