مدل هاي يادگيري ماشين براي پيش بيني تشخيص بيماري كبد

Machine learning models for predicting the diagnosis of liver disease

ﺳﺎﺑﻘﻪ و ﻫﺪف: ﮐﺒﺪ ﻣﻬﻢ ﺗﺮﯾﻦ ارﮔﺎن داﺧﻠﯽ ﺑﺪن ﻣﯽ ﺑﺎﺷﺪ ﮐﻪ ﻧﻘﺶ اﺻﻠﯽ در ﻣﺘﺎﺑﻮﻟﯿﺴﻢ ﺑـﺪن دارد. ﺑﯿﻤـﺎري ﮐﺒـﺪ را ﻧﻤﯽ ﺗﻮان ﺑﻪ راﺣﺘﯽ در ﻣﺮاﺣﻞ اوﻟﯿﻪ ﮐﺸﻒ ﮐﺮد زﯾﺮا ﮐﺒﺪ ﺣﺘﯽ زﻣﺎﻧﯽ ﮐﻪ ﻗﺴﻤﺘﯽ از آن ﻧﯿﺰ آﺳﯿﺐ دﯾﺪه ﺑﺎﺷﺪ ﺑـﻪ درﺳـﺘﯽ ﮐﺎر ﻣﯽ ﮐﻨﺪ و اﯾﻦ ﺧﻮد ﺗﺸﺨﯿﺺ اﯾﻦ ﺑﯿﻤﺎري را ﻣﺸﮑﻞ ﻣﯽ ﮐﻨﺪ. اﺑﺰارﻫﺎي ﻃﺒﻘﻪ ﺑﻨﺪي اﺗﻮﻣﺎﺗﯿﮏ ﺑﻪ ﻋﻨﻮان ﯾﮏ اﺑﺰار ﮐﻤـﮏ ﺗﺸﺨﯿﺺ ﺑﺎﻋﺚ ﮐﺎﻫﺶ ﺑﺎر ﮐﺎري ﭘﺰﺷﮑﺎن ﻣﯽ ﮔﺮدد. ﻃﺒﻘﻪ ﺑﻨﺪي ﻫﺎﯾﯽ ﮐﻪ ﺑﻪ ﻣﻨﻈﻮر ﺗﺸﺨﯿﺺ ﻫﻮﺷﻤﻨﺪ ﺑﯿﻤﺎري ﮐﺒﺪ در اﯾـﻦ ﭘﮋوﻫﺶ ﻣﻮرد اﺳﺘﻔﺎده ﻗـﺮار ﮔﺮﻓﺘـﻪ اﺳـﺖ ﺷـﺎﻣﻞ دﺳـﺘﻪ ﺑﻨـﺪﻫـﺎي ,Trees Random Forest 1NN, Naïve Bayes, SVM, AdaBoost ﻣﯽ ﺑﺎﺷﻨﺪ. ﻣﻮاد و روش ﻫﺎ: داده ﻫﺎي ﻣﻮرد اﺳﺘﻔﺎده از ﺳﻮاﺑﻖ 583 ﺑﯿﻤﺎر اﺳﺖ ﮐﻪ اﯾﻦ ﻣﺠﻤﻮﻋـﻪ داده در داﻧﺸـﮕﺎه ﮐﺎﻟﯿﻔﺮﻧﯿـﺎ در ﺳﺎل 2013 ﺑﻪ ﺛﺒﺖ رﺳﯿﺪه اﺳﺖ. ﺑﺮاي ارزﯾﺎﺑﯽ ﻣﺪل ﻫﺎي اﺳﺘﻔﺎده ﺷﺪه از اﻋﺘﺒﺎرﺳﻨﺠﯽ ﺿﺮبدري از ﻧﻮع k-ﻻﯾﻪ اﺳﺘﻔﺎده ﺷﺪه اﺳﺖ. 5 ﻣﺪل ﻣﺎﺷﯿﻦ ﯾﺎدﮔﯿﺮي از ﻧﻈﺮ وﯾﮋﮔﯽ، ﺣﺴﺎﺳﯿﺖ، ﺳﻄﺢ زﯾﺮ ﻣﻨﺤﻨﯽ راك و دﻗﺖ دﺳﺘﻪ ﺑﻨﺪي ﻣﻘﺎﯾﺴﻪ ﺷﺪﻧﺪ 0/67 ،0/59 ،0/72 و 0/5 اﺳﺖ. ﻧﺘﯿﺠﻪ ﮔﯿﺮي: ﻣﺪل Trees Random Forest ﺑﻬﺘﺮﯾﻦ ﻣﺪل ارزﯾﺎﺑﯽ ﮔﺮدﯾﺪ ﮐﻪ داراي ﺑﺎﻻﺗﺮﯾﻦ ﻣﯿﺰان دﻗﺖ ﻣﯽ ﺑﺎﺷﺪ. از ﻧﻈﺮ ﺳﻄﺢ زﯾﺮ ﻣﻨﺤﻨﯽ راك ﻣﺪل Trees Random Forest و Naïve Bayes ﺑـﯿﺶ ﺗـﺮﯾﻦ ﺳـﻄﺢ زﯾـﺮ ﻣﻨﺤﻨـﯽ را دارا ﻣﯽ ﺑﺎﺷﻨﺪ. ﻟﺬا ﺑﻪ ﮐﺎرﮔﯿﺮي ﻣﺪل Trees Random Forest در زﻣﯿﻨﻪ ﺗﺸﺨﯿﺺ و ﭘﯿﺶ ﺑﯿﻨﯽ ﺑﯿﻤﺎري ﮐﺒﺪ ﭘﯿﺸﻨﻬﺎد ﻣﯽ ﺷﻮد. اﯾﻦ اﻣﺮ در ﺗﺤﻘﯿﻘﺎت ﻣﺮﺗﺒﻂ ﺑﺎ ﺣﻮزه ي ﺳﻼﻣﺖ و ﺑﻪ ﺧﺼﻮص در ﺗﺨﺼﯿﺺ ﻣﻨـﺎﺑﻊ درﻣـﺎﻧﯽ ﺑـﺮاي اﻓـﺮادي ﮐـﻪ ﭘﺮﻣﺨـﺎﻃﺮه ﭘﯿﺶ ﺑﯿﻨﯽ ﻣﯽ ﺷﻮﻧﺪ از اﻫﻤﯿﺖ ﺑﺎﻻﯾﯽ ﺑﺮﺧﻮردار اﺳﺖ.

Introduction: The liver is the most important organ of the body has a central role in metabolism. Liver disease cannot be easily discovered in the early stages, because even when the liver is damaged partially, it also can work truly, and this makes it difficult to diagnose. Automatic classification tools as a diagnostic tool can reduce the workload of doctors. Smart ways to detect liver disease classification used in this study consist of classifier and Naïve Bayes, Trees Random Forest, 1NN, AdaBoost, SVM. Materials and Methods: Our database was 583 patient records which they have been registered at university of California in 2013. For evaluate the proposed models, it is used K-fold cross validation. Five models of machine learning compare base on specificity, sensitivity, accuracy and area under ROC curve. Results: The accuracy of the five models, respectively, 55%, 72%, 64%, 70% and 71% respectively and area under the ROC curve of 0.72, 0.72, 0.59, and 0.67 is 0.5.Conclusion: Trees Random Forest model was the best model with the highest level of accuracy. The area under the ROC curve of Trees Random Forest and Naïve Bayes models have the largest area under the curve. Therefore Trees Random Forest model and predict the diagnosis of liver disease is recommended.