Nano fuzzy alarming system for blood transfusion requirement detection in cancer using deep learning

Abstract

Periodic blood transfusion is a need in cancer patients in which the disease process as well as the chemotherapy can disrupt the natural production of blood cells. However, there are concerns about blood transfusion side effects, the cost, and the availability of donated blood. Therefore, predicting the timely requirement for blood transfusion considering patient variability is a need, and here for the first-time deal with this issue in blood cancer using in vivo data. First, a data set of 98 samples of blood cancer patients including 61 features of demographic, clinical, and laboratory data are collected. After performing multivariate analysis and the approval of an expert, effective parameters are derived. Then using a deep recurrent neural network, a system is presented to predict a need for packed red blood cell transfusion. Here, we use a Long Short-Term Memory (LSTM) neural network for modeling and the cross-validation technique with 5 layers for validation of the model along with comparing the result with networking and non-networking machine learning algorithms including bidirectional LSTM, AdaBoost, bagging decision tree based, bagging KNeighbors, and Multi-Layer Perceptron (MLP). Results show the LSTM outperforms the other methods. Then, using the swarm of fuzzy bioinspired nanomachines and the most effective parameters of Hgb, PaO2, and pH, we propose a feasibility study on nano fuzzy alarming system (NFABT) for blood transfusion requirements. Alarming decisions using the Internet of Things (IoT) gateway are delivered to the physician for performing medical actions. Also, NFABT is considered a real-time non-invasive AI-based hemoglobin monitoring and alarming method. Results show the merits of the proposed method.