|
| First author, year of publication, and country | Aim of the study | Data | ML method | Validation results | More information |
|
| Al-jaboriy et al., 2019, Malaysia [17] | ALL segmentation | Blood smear images (ALL-IDB) | ANN | Accuracy = 97% | The proposed model detected 625 cells out of 540 WBC |
|
| Al-Tahhan et al., 2020, Egypt [19] | Automatic detection ALL | Blood smear images (ALL-IDB2) | KNN
SVM
ANN | Accuracy of testing = 100%
F1-score = 100% | Quadratic SVM has the best performance in detecting ALL among ALL-IDB2 dataset |
|
| Bodzas et al., 2020, Czech [24] | Automated detection of ALL | Blood smear images (local) | SVM
ANN | Sensitivity = 100%
Specificity = 95.31% | Artificial neural network has the best performance in detecting ALL |
|
| Boldú et al., 2019, Spain [23] | Automatic recognition of different types of blast | Peripheral blood images (local) | LDA | Six groups of cell accuracy = 85% and for some class, accuracy was 97% | Classification accuracy for the six groups of cell types was 85.8 |
|
| Dasariraju et al., 2020, USA [27] | Detection and classification of immature leukocytes for diagnosis of AML | Single-cell morphological dataset of leukocytes from AML patients and nonmalignant (public) | RF | Accuracy of detection of immature = 92.99%
Accuracy for classification of immature leukocytes for types = 93.45% | Segmentation, feature extraction, detection and classification, and calculation modules were applied |
|
| Fathi et al., 2018, Egypt [16] | Classification of ALL from normal cases | Blood smear dataset (ALL-IDB) | SVM with a Gaussian radial basis kernel | Accuracy = 96.2%
Sensitivity = 97.3%
Specificity = 95.3% | Goal of this research was to design a framework for classification of cancer based on medical images |
|
| Fan et al., 2019, China [77] | Localization and segmentation | Four commonly used blood smear dataset (BCISC, LISC, and 2 other released datasets) | DNN | Dataset 1: precision = 0.995%
Dataset 2: precision = 0.994%
Dataset 3: precision = 0.989%
Dataset 4: precision = 0.984% | Proposed Leukocyte Mask architecture to gain best precision result with all datasets |
|
| Khilji et al., 2020, Bangladesh [75] | Detection of ALL | ALL dataset (C_NMC) | CNN-based different models (encrypted) | Accuracy = 77.934% | Proposed model compared with other state-of-the-art model and gain better accuracy |
|
| Nagiub Abdelsalam et al., 2018, Egypt [26] | Detection of all types of leukemia | Leukemia microscopic | CNN | Accuracy = 99.98% | Different types of pretrained (CNN) models were applied and Inception-v3 model had the highest accuracy |
|
| Nagiub Abdelsalam et al., 2019, Egypt [25] | AML detection | AML microscopic images (local) | CNN (ResNet-101) | Accuracy = 100%
Sensitivity = 100% | Even deep neural networks: AlexNet, GoogLeNet, VGG16, VGG19, Inception-v3, ResNet50, and ResNet10 |
|
| Praveena and Singh, 2020, India [78] | Segmentation and classification of ALL | ALL-IDB2 | Sparse-FCM and deep convolutional neural network | Accuracy = 93.5%
Sensitivity = 95.28%
Specificity = 93.89% | Grey Wolf-based Jaya optimization algorithm was applied for training CNN |
|
| Sharif et al., 2020, Pakistan and Qatar [76] | Recognition of different types of leukocytes | LISC, ALL-IDB1, and ALL-IDB2 malignant (public) | Localization using YOLOv2. Classification using PSO | Accuracy for ALL-IDB1 = 97.2%
Accuracy for ALL-IDB2 = 100%
Accuracy for LISC > 99% | Naïve Bayes and discriminant analysis and particle swarm optimization was used |
|
| Vogado et al., 2018, Brazil [22] | Diagnosis of ALL | ALL-IDB1
ALL-IDB2
Leukocytes
CellaVision | CNN (AlexNet + CaffeNet + Vgg-f) and SVM | Accuracy = 100%
Precision = 100% | Goal of this research was to design a framework for classification of cancer based on medical images (3 architectures were used in feature extraction, SVM for classification) |
|
| Shafique and Tehsin, 2020 Pakistan [79] | Detection and classification of ALL | ALL-IDB1 and ALL-IDB2 | CNN (AlexNet) | ALL detection accuracy = 99.50%, ALL subtype classification = 96.06%, dataset precision = 0.984% | After detection, ALL subtype was classified based FAB classification system. Datasets |
|
| Hegde et al., 2018, India [18] | Detection of nuclei and classification of WBC | Leishman | SVM | Accuracy of detection of lymphocyte = 100% | After segmentation, the nucleus of WBC cells different kinds of them was classified |
|
| Jha and Dutta, 2019, India [21] | Detection of ALL | ALL-IDB2 | Proposed hybrid segmentation + Chrono-SCA-ACNN | Accuracy = 99%
Sensitivity = 100% | Nucleus and cytoplasm segmentation using Chrono-SCA-ACNN (chronological sine cosine algorithm-based actor-critic neural network) |
|
