Miyelodisplastik sendromu hastalığında erken ve ayırıcı teşhis için makine öğrenmesi yöntemlerinin performans kıyaslaması ve hibrit bir model önerisi
Küçük Resim Yok
Tarih
2023
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
İstanbul Beykent Üniversitesi
Erişim Hakkı
info:eu-repo/semantics/openAccess
Özet
Miyelodisplastik Sendrom (MDS) tanısı zor bir tür kemik iliği hastalığıdır. MDS' nin kesin nedeni bilinmemektedir ve tanısı geciktiğinde "Akut Lösemi" olarak tanımlanan hızlı başlangıçlı ve agresif lösemilere dönüşebilmektedir. Hastalığın tanısı aşamasında alanında uzman bir Hematolog tarafından klinik bulgular, patolojik bulgular, laboratuvar bulguları birleştirilerek elde edilen veriler, uzman deneyimleri ile yorumlanarak şüpheli bir durum var ise ayırıcı tanı için gerekli tetkikler ve bunların sonucunda da ek değerlendirmeler ile sonuca ulaşılabilmektedir. Bu süreç hem hasta ile hasta yakını hem de sağlık sistemi için maddi, manevi zorlayıcı bir süreç olmaktadır. İnsan gözüyle ayırt edilemeyen ya da o anda önemsiz görülen her küçük ayrıntı bir makine öğrenmesi algoritmaları ile yakalanıp, işlenebilmektedir. Bu çalışmada, Bursa Uludağ Üniversitesi Hastanesi'nden alınan MDS tanısı olan ve MDS tanısı olmayan kişilerin yaş, cinsiyet, halsizlik, hemoglabin gibi klinik belirti, bulgu ve tetkik verileri ele alınarak makine öğrenmesi yöntemlerinden olan Lojistik Regresyon (LR), Rastgele Orman (RO), XGBoost, Çok Katmanlı Algılayıcılar (ÇKA), Naive Bayes(NB) ve önerilen hibrit bir yöntem ile modeller oluşturulmuş, MDS erken ve ayırıcı teşhis için kullanılabilirliği analiz edilmiş ve performansları değerlendirilmiştir. Yapılan çalışmalar sonucunda geliştirilen hibrit bir modelin %85 doğruluk ile diğer makine öğrenmesi yöntemlerine göre daha iyi sonuç verdiği ortaya konulmuştur.
Myelodysplastic Syndrome (MDS) is a type of bone marrow disease. The exact cause of MDS is unknown, and when the diagnosis is delayed, it can turn into rapid-onset and aggressive leukemia defined as "Acute Leukemia". When the diagnosis stage of the disease is passed, the clinical findings, pathological findings, and laboratory findings are combined by a hematologist, and data are obtained. By interpreting in line with the results reached by the expert's experience, if there is a suspicious situation, the correct result is reached with the necessary tests for differential diagnosis and additional evaluations as a result of these. This stage is both a financial and moral difficult process for the patient, their relatives and the health system. Every little detail that is indistinguishable to the human eye or considered unimportant at that moment can be captured and processed by machine learning algorithms. In this study, Logistic Regression (LR), Random Forest (RF), XGBoost, Multiple Layer Perceptron (MLP), Naive Bayes (NB) and a proposed hybrid method value was used by using the considering the clinical signs, findings and examination data such as age, gender, fatigue, hemoglobin of people with and without MDS diagnosis from Bursa Uludağ University Hospital and their usability for early and differential diagnosis of MDS was analyzed and their performance evaluated. As a result of the studies, it has been revealed that a hybrid model developed with 85% accuracy gives better results than other machine learning methods.
Myelodysplastic Syndrome (MDS) is a type of bone marrow disease. The exact cause of MDS is unknown, and when the diagnosis is delayed, it can turn into rapid-onset and aggressive leukemia defined as "Acute Leukemia". When the diagnosis stage of the disease is passed, the clinical findings, pathological findings, and laboratory findings are combined by a hematologist, and data are obtained. By interpreting in line with the results reached by the expert's experience, if there is a suspicious situation, the correct result is reached with the necessary tests for differential diagnosis and additional evaluations as a result of these. This stage is both a financial and moral difficult process for the patient, their relatives and the health system. Every little detail that is indistinguishable to the human eye or considered unimportant at that moment can be captured and processed by machine learning algorithms. In this study, Logistic Regression (LR), Random Forest (RF), XGBoost, Multiple Layer Perceptron (MLP), Naive Bayes (NB) and a proposed hybrid method value was used by using the considering the clinical signs, findings and examination data such as age, gender, fatigue, hemoglobin of people with and without MDS diagnosis from Bursa Uludağ University Hospital and their usability for early and differential diagnosis of MDS was analyzed and their performance evaluated. As a result of the studies, it has been revealed that a hybrid model developed with 85% accuracy gives better results than other machine learning methods.
Açıklama
Anahtar Kelimeler
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrol, Computer Engineering and Computer Science and Control
