Segmentation of Mammogram Abnormalities Using Ant System Based Contour Clustering Algorithm

2022 TAGA Journal  
The Computer-Aided Detection Systems (CADs) can locate and identify the normal and pathological tissues in mammogram images by segmentation. The existing segmentation methods have to test each and every pixel of the image at least once, which is computationally expensive. Objective: This research focuses on detection of microcalcifications from the digital mammograms by segmentation, where the abnormal tissues are segmented from the normal tissues. Methods: To detect microcalcifications from
more » ... digital mammograms by segmentation, a novel segmentation approach based on Ant Clustering method namely Ant System based Contour Clustering (ASCC), simulates the ants' foraging behavior is proposed. The proposed ASCC is compared with the state-of art existing methods with respect to area, pixel and edge based metrics on the Mammographic Image Analysis Society (MIAS) Dataset. Results: The segmentation performance of the proposed ASCC method is experimented on 312 digitized mammogram images acquired from the 161 patient's left and right breast Screening. The segmentation by the proposed ASCC is evaluated by the area, pixel and edge based metrics shows that 62.47% common area between overlapping segmented and the reference region by Jaccard index, Goodness based on inter-region contrast of 66.59%, Low Segmentation Error of 9.51%, precision of 93.67%, Recall of 90.90%, 0.85% Figure of Merit, Over-segmented Pixel Rate of 0.43%, and Under-Segmented Pixel Rate of 0.26%. Conclusion: Segmentation is key preprocessing method to accurately locate and identify the normal and pathological tissues in digital mammogram images. This study proposes an ASCC method for segmentation task by hybridizing clustering and contour based segmentation approaches. The evaluated results with respect to area, pixel and edge based metrics shows added advantage in segmentation tasks compared to the other approaches.
doi:10.37896/pd91.4/91426 fatcat:qrs327mc5zhebgsmk2vrgvcqeu