Table 1 A brief comparison of the current leukemia classification methods.

Ramaneswaran et al.¹

A new Hybrid Model (HM) was proposed for ALL classification. HM combines between Inception v3 and XGBoost. Inception v3 was used for feature extraction process. While XGBoost was used for classification

The proposed model demonstrates a high degree of accuracy and reliability in detecting acute lymphoblastic leukemia cells

The used dataset is limited

Ullah et al.¹⁷

A diagnostic support system that utilizes a CNN and an ECA module was used to accurately classify images of cancerous and healthy cells. The VGG16 framework was utilized to extract features from the source images. The ECA module was added after each convolutional block to enhance the importance of the extracted features from VGG16

Incorporating the attention module into deep learning architectures can result in a substantial improvement in performance

The proposed model suffers from high false positive rate

Das and Meher¹⁸

An innovative deep CNN framework was presented to enhance the efficiency of detecting and classifying ALL. A new method was suggested that incorporates a probability-based weight factor to merge MobilenetV2 and ResNet18 efficiently, while still benefiting from the strengths of both techniques

The proposed framework overcome the problem of required huge dataset

The proposed model suffers from time complexity

Jawahar et al.¹⁹

A new model which is called ALNett was introduced. The custom CNN architecture utilizes a series of stacked convolution layers to acquire hierarchical features, resulting in precise classification. The inclusion of batch normalization between these stacked clusters improves the extracted features, leading to improved classification performance. The normalization process improves the weights and learning process, leading to a substantial reduction in features within the stacked hierarchical clusters. As a result, all microscopic images can be classified accurately and quickly

ALNett exhibited encouraging performance in categorizing ALL and surpassed the performance of the other pre-trained models

The system is very complex

Saeed et al.²⁰

CNN-based adaptation model DeepLeukNet classifies ALL using microscopy images. Data augmentation has been used to generate additional images to reduce model overfitting. After visualizing intermediate layer, ConvNet filter, and heatmap layer activation, a qualitative analysis was performed. The proposed model was also compared to existing methods to verify its efficacy

The proposed model provides high accuracy

The proposed model suffers from time complexity

Mallick et al.²¹

A new training Deep Neural Networks (DNN) has been proposed. Actually, the proposed DNN aims to classify leukemia types. Due to the substantial size of the gene expression data, a neural network with seven layers is developed to accurately classify two distinct forms of leukemia. The classification accuracy is significantly superior in comparison to other types of classifiers. The user’s text is very short and does not provide any information

DL models possess the ability to process and analyze extensive quantities of data, subsequently extracting significant features without human intervention

The proposed DNN doesn’t give the optimum accuracy

Alzahrani et al.²²

A new algorithm that can distinguish between and classify two distinct forms of leukemia: ALL and AML has been introduced. The three main parts of the algorithm are image preprocessing, segmentation, and classification. The segmentation feature is complete. Classification procedures are carried out with the help of a newly constructed UNET that utilizes a U-shaped architecture

The proposed system can enhance the capabilities of healthcare providers in delivering their services

One significant obstacle faced during this research was the absence of all types of leukemia in the datasets, which were limited to ALL and AML only. Therefore, these datasets pose various challenges. Only two categories have varying dimensions, necessitating input resizing