Browsing by Author "Nikolic, Milos (57224348525)"

Purpose: Radioactive iodine therapy (RAIT) is important when treating patients who have been diagnosed with differentiated thyroid carcinoma and have gone through initial surgery. However, deciding whether a patient should undergo such therapy as well as the proper iodine dose is a complex task, especially for those with a lack of experience. Therein, this paper aimed to develop and compare classifier systems to aid inexperienced physicians in decision making on radioactive iodine therapy for thyroid cancer patients. Methods: The study cohort consisted of 210 thyroid cancer patients who had undergone a total thyroidectomy. We developed and evaluated the performance of three machine learning (ML) algorithms that suggest whether these patients should undergo RAIT and propose an administrable I-131 dose. These algorithms were Artificial Neural Network (ANN), Naïve Bayes Classifier (NB) and Group Method of Data Handling (GMDH). The kappa coefficient was used to measure agreement of classifiers with gold standard decision made by an experienced physician. Results: Our results indicate that the ANN performs better than NB and GMDH in terms of accuracy (95.71%). On the basis of the Kappa coefficient, ANN was also the best 0.96 (0.91-1.00). Additionally, kappa coefficient increased to 0.93 (0.86-1.00) by comparing young physicians' decisions on thyroid cancer therapy before and after using ANN as a decision making tool. Conclusion: Our results suggest that developed classifiers are able to imitate the real decisions of medical expert. Furthermore, classifiers may be utilized to educate inexperienced medical professionals, especially in the absence of strict guidelines' recommendations. © 2021 Zerbinis Publications. All rights reserved.

Purpose: Radioactive iodine therapy (RAIT) is important when treating patients who have been diagnosed with differentiated thyroid carcinoma and have gone through initial surgery. However, deciding whether a patient should undergo such therapy as well as the proper iodine dose is a complex task, especially for those with a lack of experience. Therein, this paper aimed to develop and compare classifier systems to aid inexperienced physicians in decision making on radioactive iodine therapy for thyroid cancer patients. Methods: The study cohort consisted of 210 thyroid cancer patients who had undergone a total thyroidectomy. We developed and evaluated the performance of three machine learning (ML) algorithms that suggest whether these patients should undergo RAIT and propose an administrable I-131 dose. These algorithms were Artificial Neural Network (ANN), Naïve Bayes Classifier (NB) and Group Method of Data Handling (GMDH). The kappa coefficient was used to measure agreement of classifiers with gold standard decision made by an experienced physician. Results: Our results indicate that the ANN performs better than NB and GMDH in terms of accuracy (95.71%). On the basis of the Kappa coefficient, ANN was also the best 0.96 (0.91-1.00). Additionally, kappa coefficient increased to 0.93 (0.86-1.00) by comparing young physicians' decisions on thyroid cancer therapy before and after using ANN as a decision making tool. Conclusion: Our results suggest that developed classifiers are able to imitate the real decisions of medical expert. Furthermore, classifiers may be utilized to educate inexperienced medical professionals, especially in the absence of strict guidelines' recommendations. © 2021 Zerbinis Publications. All rights reserved.