Deep Learning–Based Detection and Stage Classification of Blood Cancer from Microscopic Smear Images and Numerical Risk Factors

Abstract

Blood cancer, particularly leukemia, is a critical hematological malignancy that originates in the bone marrow and blood and can spread rapidly throughout the body. Patients commonly present with non-specific symptoms such as persistent fatigue, fever, recurrent infections, unexplained weight loss, night sweats, shortness of breath, bone and joint pain, enlarged lymph nodes, and abnormal bleeding or bruising. These signs are often overlooked or confused with common illnesses. Risk may be influenced by genetic mutations, family history of hematological disease, exposure to ionizing radiation or toxic chemicals such as benzene, smoking, and certain prior chemotherapy or radiotherapy. Because of the non-specific symptoms and complex risk profile, many cases are diagnosed at an advanced stage, highlighting the need for early, objective diagnostic support. This paper proposes a deep learning–based computer-aided diagnosis system for automatic analysis of blood cancer using both microscopic images and numerical data. Peripheral blood smear images are preprocessed using a median filter to remove noise while preserving essential morphological  details. A Convolutional Neural Network (CNN) extracts discriminative spatial features, which are combined with anonymized numerical parameters such as complete blood count values, basic clinical details, and selected lifestyle and food habit indicators. An Artificial Neural Network (ANN) classifier then uses these fused features for cancerous/non-cancerous detection and stage-wise classification into early, intermediate, and advanced categories, supporting more accurate diagnosis and risk assessment.