To detect chronic kidney disease (CKD) at earlier stages, diagnosis through non-invasive ultrasonographic imaging techniques provides an auxiliary clinical approach for at-risk CKD patients. We have established a detection method based on imaging processing techniques and machine learning approaches for the diagnosis of different CKD stages. Decisive area-proportional and textural features and support-vector-machine techniques were applied for efficient and effective analyses. Several clustered collections of CKD patients were evaluated and compared according to the estimated glomerular filtration rates. Based on the findings of evolving changes from ultrasound images, the proposed approach could be used as complementary evidences to help differentiate between different clinical diagnoses.