Leukemia is one of the deadliest diseases that occur in white blood cells which is identified in microscopic images from blood samples. Researchers have developed various techniques to diagnose leukemia using a machine learning approach by analyzing micro imaging blood samples. During image screening, the Damaged cells be identified based on manually counting on similar structure lymphocytes and monocytes structures, but the variant difference of both cells aren’t identified accurately. So the identification failures that occur lead to failed accurate cancer detection because feature variants project equal cell structure. To resolve this problem, we propose an Invariant Structural Cascade Segmentation (ISCS) based on Deep Vectorized Scaling Neural Network (DVSNN) is implemented to detect the Leukemia Cancer automatically from bio-blood samples micro image for early diagnosis. First, the Leukemia Cancer micro-image dataset (LCMID) was collected and preprocessed into a noise-free dataset based on adaptive median filters. Then the segmentation was carried to partition the microcells using Invariant structural cascade segmentation (ISCS) optimized with watershed to identify the features such as texture, pixel color, pixel intensity. For identifying the structural components of cells variation difference using Angular Vector Projection (AVP) was applied to find the structural variance. Then the histogram color equalizer (HCE) was applied to select the damaged cells using K-counts and their feature weight. Then the selected feature weights are trained into Deep vectorized scaling the neural network to classify the risk of the Leukemia cancer cells. This identifies the cancer cells effectively from micro image cells for detecting the risk of the patients. This improves the sensitivity, specificity rate as well in classification accuracy compared to the other methods.