A Pyramid Model of Convolutional Neural Network to Classify Acute Lymphoblastic Leukemia Images

Abstract

Many researchers have classified acute lymphoblastic leukemia using several methods. One of the methods is a convolutional neural network. However, the limitation of the convolutional neural network is a large number of trainable parameters updated. The paper proposes a new architecture based on the convolutional neural network. We have designed and implemented a convolutional neural network with different kernels, where we increase the number of kernels like pyramid models. We utilized the final convolution to conduct the fully connected, followed by the SoftMax function to classify the image. We have evaluated our proposed architecture using acute lymphoblastic leukemia image database (ALL-IDB2). The results show that our proposed method produced the average Accuracy, Precision, and Recall of 99.17%, 99.33%, and 99%, respectively. It has outperformed other models, i.e., shape features, Multi distance of GLCM, CNN and SVM, Pretrained deep convolutional neural networks, AlexNet, ensemble network, convolutional and recurrent neural network, and hypercomplex-valued convolutional neural networks.

A New Model: Commutative Hypercomplex – Convolutional Neural Network to Classify Acute Lymphoblastic Leukemia Images

“…We also compare the maximum accuracy of our proposed method to the others. Our proposed method is better than the others, except the Pyramid model [20]. The pyramid model has generated the same accuracy as our proposed method for A[+1, -1] and A[-1,+1] commutative hypercomplex models.…”

“…Moreover, the Pyramid model [20] and hypercomplex model [18] have produced better accuracy than our proposed method. However, It is not better than the multi distance model [3], Pyramid model [20], and hypercomplex model [18]. In addition, our proposed method A[-1, +1] commutative hypercomplex model has exceeded the accuracy of A[+1, -1] commutative hypercomplex model, the multi distance model [3] Hypercomplex model [18].…”

“…Fig. 19 shows that our proposed method A[+1, -1] Commutative Hypercomplex model outperformed KNN [7], SVM-RBF [7], SVM-L [7], SVM-P [7], Naïve Bayes Gaussian [7], Naïve Bayes complement [7], decision tree [7], colour hybrid modelling [20]. Moreover, the Pyramid model [20] and hypercomplex model [18] have produced better accuracy than our proposed method.…”

“…In addition, our proposed method A[-1, +1] commutative hypercomplex model has exceeded the accuracy of A[+1, -1] commutative hypercomplex model, the multi distance model [3] Hypercomplex model [18]. However, it is not better than Pyramid model [20].…”

“…Furthermore, we compare our proposed method results to the others, such as K-nearest neighbour (KNN) [7], support vector machine-radial basis function (SVM-RBF) [7], support vector machinelinear (SVM-L) [7], support vector machinepolynomial (SVM-P) [7], Naïve Bayes Gaussian [7], Naïve Bayes complement [7], decision tree [7], Muti distance model [3], colour hybrid modelling pyramid model [20], and hypercomplex model [18] as shown in Fig. 19.…”