Article Preview
Top1. Introduction
Touted as the leading cause of “cancer” death across the globe, lung cancer has been among the most common type of malignancies diagnosed on adults (Ferlay, et al., 2010; Sigel, et al., 2020). Aiding decisions in operative, perioperative, and/or surgical thoracic procedures, researchers such as Desuky & El Bakrawy (2016) and Danjuma (2015) have evaluated the performance of machine learning (ML) algorithms, for example, multilayer perceptron (MLP), J48, and the Naive Bayes (NB), on the University of California Irvine (UCI) Machine Learning (ML) repository thoracic surgery dataset.
As Figure 1 shows, thoracic surgery may be split into three (3) specialties: (1) adult cardiac surgery; (2) congenital or pediatric heart surgery; and (3) general thoracic surgery.
Figure 1.
Consequence of Thoracic Surgery about here
In operative, perioperative, and/or surgical critical care of patients (American Medical Association, n.d.) who obtained congenital pathologic conditions within the chest, thoracic surgery is often recommended. Even so, recent studies have predicted that around 80% of lung cancer patients are diagnosed with non-small cell lung cancer (NSCLC) and around 25% with early-stage operable disease (e.g., Adam, et al., 2014; Timmerman, et al., 2016; Sarna, et al, 2008). For the NSCLC early stages, the preferred treatment has been curative lung resection. Symptoms for NSCLC include pain, fatigue, the decay of lung function, cardiorespiratory fitness and quality of life. Similar to NSCLC, small-cell lung cancer (SCLC) is very aggressive.
Notwithstanding, with the low post-surgery survival rate of lung cancer patients whether it is SCLC or NSCLC (e.g., Timmerman et al, 2016; Adam et al, 2019), many critical factors such as age, experience of the surgeon and patient medical condition, among other things, must be considered in determining the risk of operating on these patients. Hence, a thorough diagnosis and analysis must be performed based on past historic patient data and current patient medical condition prior to recommending surgery. With respect to the prediction of post-thoracic life expectancy, there has been emerging work implementing ML techniques. Predictions from the use of these techniques are often good enough to assist the patient (and surgeon) in deciding to undergo surgery or not.
Given the limited cancer patient survival rate post-thoracic surgery, research has emerged in applying data mining techniques for its medical diagnosis and prediction (e.g., Nachev & Reapy, 2015). Models used included decision trees, Naïve Bayes (NB), artificial neural network (ANN) and support vector machines (SVM). More recently, Desuky & El Bakrawy (2016) have applied MLP, Naïve Bayes, J48, logistic regression (LR) for post-thoracic surgery life expectancy prediction on the UCI thoracic surgery dataset. Their work also involved attribute ranking and selection to achieve more accurate prediction. While only satisfactory accuracy has been achieved with traditional ML algorithms as well as with more recent ones such as Multi-Layer Perceptron (MLP), a type of Artificial Neural Network (ANN), and Bayesian model, no work on deep learning (DL) for post-thoracic life expectancy has yet been found.