Prior methods had trouble developing an effective segmentation framework model for detecting liver tumors from CT images, leading to subpar results in terms of both visual appeal and morphological accuracy. Scale and focus changes almost always have an effect on abnormal liver CT image sets, in contrast to statistical measures. As such, an effective ACM model is developed to partition the liver areas, accounting for boundary discontinuity and variations. Parametric in nature, the presented algorithms take advantage of initial parameter sets to achieve better results. In order to adapt to shifting liver boundaries and increase precision, we present a gradient-driven ROI model with automatic texture and shape feature extraction.