Machine Learning and Feature Selection Applied to SEER Data to Reliably Assess Thyroid Cancer Prognosis

Abstract

Utilizing historical clinical datasets to guide future treatment choices is beneficial for patients and physicians. Machine learning and feature selection algorithms (namely, fisher's discriminant ratio, Kruskal-Wallis' analysis, and Relief-f) have been combined in this research to analyse a SeeR database containing clinical features from de-identified thyroid cancer patients. The data covered 34 unique clinical variables such as patients' age at diagnosis or information regarding lymph nodes, which were employed to build various novel classifiers to distinguish patients that lived for over 10 years since diagnosis, from those who did not survive at least five years. By properly optimizing supervised neural networks, specifically multilayer perceptrons, using data from large groups of thyroid cancer patients (between 6,756 and 20,344 for different models), we demonstrate that unspecialized and existing medical recording can be reliably turned into power of prediction to help doctors make informed and optimized treatment decisions, as distinguishing patients in terms of prognosis has been achieved with 94.5% accuracy. We also envisage the potential of applying our machine learning strategy to other diseases and purposes such as in designing clinical trials for unmasking the maximum benefits and minimizing risks associated with new drug candidates on given populations. Machine learning as algorithmic advancement in the past few years dramatically improved our range of potential implementation of artificial intelligence for tasks such as learning and playing the Go game, environment feature recognition for self-driving, and in medical applications 1,2. Within the machine learning scope, artificial neural networks (ANNs) are a set of algorithms that recognize patterns and learn from inputs and outputs to make useful connections without pre-set rules 3. Furthermore, ANNs and their performance correlate well with the training data size and are more adept at pattern recognition and classification when analysing large hospital records than traditional statistical modelling applied in some of the more recent cancer prognostication applications 4,5. ANN models are designed in layers to learn increasingly higher-dimension and remote representations of the input data and devise meaningful outcomes to feed the next layer. In this work, we tested three separate neural network models to determine the outcomes of thyroid cancer patients after diagnosis from distilling the U.S. Surveillance Epidemiology and End Results (SEER) database. Although back in 2015 thyroid cancer cases in the United States were predicted to increase to 92,000 by 2020 6 , and current estimates indicate that in 2019 around 52,000 are projected instead, these numbers still signify that thyroid cancer incidence rates continue to increase 7. Specifically, regarding women, thyroid cancer ranks sixth compared to other types of cancer in terms of incidence with almost 38,000 new estimated cases per year 7. These trends can be mainly attributed to an increase in ...

Application of Data Science Approaches to Investigate Autoimmune Thyroid Disease in Precision Medicine

“…Data science is known to encompass the preparation of data for analysis, this includes aggregating, cleaning, and manipulating the data to uncover patterns and draw out insights. Exploiting historical clinical datasets to improve future treatment choices has proved beneficial for both patients and physicians [43,51]. Through machine learning (a branch of artificial intelligence), it is very possible to obtain patterns within patient data, the exploitation of these patterns helps to predict and treat patients in order to improve clinical disease management [52].…”

“…Machine learning also features selection algorithms such as Kruskal-Wallis' analysis, Fisher's discriminant ratio, and Relief-F. In some research, these algorithms have been used to analyze databases containing clinical features (such as U.S. Surveillance Epidemiology and End Results (SEER) database) from identified thyroid disease patients [51].…”

Harnessing the Potential of Human Pluripotent Stem Cell-Derived Motor Neurons for Drug Discovery in Amyotrophic Lateral Sclerosis: From the Clinic to the Laboratory and Back to the Patient

“…Taking advantage of databases like the PRO-ACT database, together with genetic information and preclinical data, coupled with artificial intelligence (AI)-based strategies, it will be possible to develop novel artificial neural network-based systems to confidently predict patient survival and stratify ALS patients for enrolment in clinical trials, perhaps, improving the efficiency of the drug discovery process (Atassi et al, 2014;Ko et al, 2014;Zach et al, 2015;Zhou and Manser, 2020). For example, similar systems have already been employed clinically in the estimation of survival prognosis in patients diagnosed with eye melanoma (Damato et al, 2008), thyroid cancer (Mourad et al, 2020), glioblastoma multiforme (Hao et al, 2018), among other diseases (Kourou et al, 2015;Zhu et al, 2020). Indeed, AI-based technologies can be wisely applied to compile, digest and interpret "hidden knowledge" in large datasets and make the data usable to researchers in the ALS field and beyond (Ko et al, 2014;Kusumoto and Yuasa, 2019;Cota-Coronado et al, 2020;Zhou and Manser, 2020).…”

Using Embedded Feature Selection and CNN for Classification on CCD-INID-V1—A New IoT Dataset

“…The embedded feature selection scheme has been preferred over the filter and wrapper methods [56][57][58], and has seen success in fields such as bioinformatics [59,60], and medical research [61][62][63][64], but remains relatively new in the field of IoT security.…”