Blood cancer leukemia poses a serious threat to life and needs quick diagnosis to treat it well. Doctors look at blood samples under microscopes to spot it, but this takes a long time and can lead to errors. This paper proposes a hyper tuned LSTM for automated blood cancer detection from microscopic blood smear images. The system learns from a bunch of blood cell images that have been labeled, and it gets good at distinguishing normal cells and cancerous cells. Tests show that LSTM work well to spot blood cancer making diagnosis faster and more accurate. Our model is trained on a dataset of labeled microscopic blood images, achieving an accuracy of 96%. This system demonstrates high efficiency, reducing diagnostic errors and assisting medical professionals in early blood cancer detections.