تكامل تفاضلي با تركيب شبكه عصبي براي استخراج قوانين از پايگاه داده هاي پزشكي

A Combined Differential Evolution and Neural Network Approach to Extract Rules from Medical Databases

اﻣﺮوزه در ﻋﻠﻢ ﭘﺰﺷﮑﯽ ﺳﯿﺴﺘﻢﻫﺎي ﭘﺸﺘﯿﺒﺎﻧﯽ ﺑﺮاي ﺗﺼﻤﯿﻤﺎت ﺑﺎﻟﯿﻨﯽ ﺗﻌﺮﯾﻒ ﺷﺪه اﻧﺪ ﮐﻪ ﺑﺎ اﺳﺘﻔﺎده از دو ﯾﺎ ﭼﻨﺪ آﯾﺘﻢ از وﯾﮋﮔﯽ ﻫﺎي ﺑﯿﻤﺎر، ﭘﯿﺸﻨﻬﺎد ﻫﺎي ﻣﻔﯿﺪي اراﺋﻪ ﻣﯽدﻫﻨﺪ. اﯾﻦ ﭘﯿﺸﻨﻬﺎد ﻫﺎﺑﻪ ﻣﺘﺨﺼﺺ در ﺗﺸﺨﯿﺺ ﺑﯿﻤﺎري ﯾﺎ روﻧﺪ درﻣﺎن ﮐﻤﮏ ﻣﯽﮐﻨﻨﺪ. ﻃﺒﻘﻪﺑﻨﺪي دادهﻫﺎ در اﯾﻦ ﺣﻮزه ﻧﯿﺎزﻣﻨﺪ ﺷﻔﺎﻓﯿﺖ و اراﺋﻪي ﻋﻠﺖ ﺑﺮاي ﺗﺨﺼﯿﺺ دادهﻫﺎ ﻣﯽﺑﺎﺷﺪ. ازاﯾﻦ رو ﻻزم اﺳﺖ ﻋﻼوه ﺑﺮ ﻃﺒﻘﻪ ﺑﻨﺪي درﺳﺖ دادهﻫﺎي ﭘﺰﺷﮑﯽ، ﻋﻠﺖ ﺗﺨﺼﯿﺺ ﻫﺮ داده ﺷﺮح داده ﺷﻮد. در اﯾﻦ ﺗﺤﻘﯿﻖ، ﯾﮏ روش ﺟﺪﯾﺪ ﺑﺮ ﭘﺎﯾﻪ ي ﺗﮑﺎﻣﻞ ﺗﻔﺎﺿﻠﯽ ﺑﺮاي ﻃﺒﻘﻪ ﺑﻨﺪي ﺧﻮدﮐﺎر آﯾﺘﻢ ﻫﺎ در ﭘﺎﯾﮕﺎه داده ﻫﺎي ﭘﺰﺷﮑﯽ ﻣﻮردﻧﻈﺮ اﺳﺖ. ﺑﺮ اﺳﺎس آن، ﯾﮏ اﺑﺰار ﺑﻪ ﻧﺎم DEREx ﻣﻌﺮﻓﯽ ﺷﺪه اﺳﺖ ﮐﻪ ﺑﻪ ﻃﻮر ﺧﻮدﮐﺎر، داﻧﺶ واﺿﺢ را از ﭘﺎﯾﮕﺎه داده، ﺑﻪ ﺷﮑﻠﯽ از ﻗﻮاﻧﯿﻦ IF-THEN ( اﮔﺮ - ﺳﭙﺲ) ﺷﺎﻣﻞ ﺷﺮط ﻫﺎي ﻣﺘﺼﻞ ﺑﻪ AND و ﺑﺮ روي ﻣﺘﻐﯿﺮﻫﺎي ﭘﺎﯾﮕﺎه داده، اﺳﺘﺨﺮاج ﻣﯽ ﮐﻨﺪ. ﻫﺮ ﻓﺮد از ﺗﮑﺎﻣﻞ ﺗﻔﺎﺿﻠﯽ ﯾﮏ ﻗﺎﻧﻮن ﺑﺮاي ﻃﺒﻘﻪﺑﻨﺪي ﻣﺤﺴﻮب ﻣﯽ-ﺷﻮد. ﻣﺠﻤﻮﻋﻪ اﻓﺮاد ﺗﮑﺎﻣﻞ ﺗﻔﺎﺿﻠﯽ ﻣﺠﻤﻮﻋﻪاي از ﻗﻮاﻧﯿﻦ ﻣﯽﺑﺎﺷﻨﺪ ﮐﻪ از ﺗﻤﺎم آن ﻫﺎ ﺑﺮاي ﻃﺒﻘﻪﺑﻨﺪي دادهﻫﺎي ﻣﻮﺟﻮد در ﭘﺎﯾﮕﺎه داده اﺳﺘﻔﺎده ﻣﯽﺷﻮد. اﯾﻦ ﻗﻮاﻧﯿﻦ ﻣﯽ ﺗﻮاﻧﻨﺪ ﺑﻪ ﺻﻮرت ﻣﻨﻄﻘﯽ ﻣﺘﺼﻞ ﺑﻪ OR ﯾﺎ ﻧﯿﺰدر ﻧﻈﺮ ﮔﺮﻓﺘﻪ ﺷﻮﻧﺪ؛ ﺑﻨﺎﺑﺮاﯾﻦ ﺗﻤﺎم ﻗﻮاﻧﯿﻦ ﻃﺒﻘﻪ ﺑﻨﺪي ﺑﺮاي ﺗﻤﺎم دﺳﺘﻪ ﻫﺎ ﯾﮏ ﺑﺎره در ﯾﮏ ﻣﺮﺣﻠﻪ ﭘﯿﺪا ﻣﯽ ﺷﻮﻧﺪ. اﺑﺰار DEREx اوﻟﯿﻦ اﺑﺰار ﻃﺒﻘﻪ ﺑﻨﺪﯾﺎﺳﺖ ﮐﻪ ﺑﺮ ﭘﺎﯾﻪ ي ﺗﮑﺎﻣﻞ ﺗﻔﺎﺿﻠﯽ ﻣﯽ ﺑﺎﺷﺪ و ﺑﻪ ﻃﻮر ﺧﻮدﮐﺎر و ﺑﺪون دﺧﺎﻟﺖ ﺳﺎزوﮐﺎر دﯾﮕﺮي دﺳﺘﻪ ﻫﺎﯾﯽ از ﻗﻮاﻧﯿﻦ IF-THEN را اﺳﺘﺨﺮاج ﻣﯽ ﮐﻨﺪ.

Today, clinical decision support systems have been defined in medical science to offer useful suggestions using two or more items of the patient's characteristics. These suggestions help specialists in disease diagnosis or in the treatment process. The classification of data in this area requires transparency and reasoning for the allocation of data. Therefore, in addition to the correct classification of medical data, it is necessary to explain the reason for the data allocation. This research uses a new method based on differential evolution for automatic classification of items in the medical database. This method has introduced a tool called DEREx, which automatically extracts explicit knowledge from the database in the form of an IF-THEN rule, including the conditions connected by AND (and) on the database variables. Each item of differential evolution is considered a rule for classification. A set of differential evolution items is a set of rules, all of which are used to classify data in the database. These rules can also be logically linked by OR (or), so all classification rules are found for all categories at once in one step. The DEREx tool is the first classification tool based on differential evolution which automatically extracts categories of IF-THEN rules without the involvement of another mechanism.