چكيده به لاتين
Breast cancer, lung cancer, skin cancer and blood malignancies such as leukemia and lymphoma are examples of different types of cancer. Cancers are a collection of cells that multiply uncontrollably in the body. Acute lymphoblastic leukemia is one of the most important forms of malignancy that occurs due to excessive production of lymphoblasts in the bone marrow.
The accurate classification of ALL subtypes is very important for its effective treatment and management, but often in the process of diagnosing leukemia, some blood malignancies are inadvertently ignored by hematologists. To prevent such unwanted mistakes, it is necessary to spend a lot of time and care. For this reason and in order to speed up this process, in this research we intend to present a new method for the classification of leukemia with the help of modern technologies such as deep learning and Neural Networks.
A study covering the period from 2006 to 2014 reported an average annual incidence rate of ALL at 2.25 per 100,000 children under 15 years of age. This rate was notably higher in males, with an annual percentage change of 7.1%, indicating an increasing trend in cases over time. The peak incidence occurred in children aged 2-5 years. In southeastern Iran, ALL is recognized as the most common hematologic malignancy, with a reported incidence of 44,000 cases of leukemia in 2008.
The research policy in this paper is divided into several interconnected parts, including creating a suitable dataset, extracting features from this dataset using a neural network architecture on each image of blood cells, and finally, classifying these images using classifiers. Deep learning is focused. The data set in this research project is first divided into two general categories, benign and malignant, then the malignant category itself is divided into three more detailed classes: early Pre-B, Pre-B and Pro-B.
This thesis presents a comprehensive study on the application of deep learning methods, including Convolutional Neural Networks (CNN), DenseNet, EfficientNet and ConvNetBase for multi-class classification of ALL. Using the power of these advanced architectures, our goal is to increase diagnostic accuracy and facilitate and accelerate therapeutic interventions. The proposed models are evaluated on a dataset of ALL, showing significant improvements in classification performance compared to other approaches.
