الگوریتم برای بهبود بهره وری تشخیصی با تنظیم مقادیر قطع شده چند منظوره نشانگرهای توموری چندگانه
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|4573||2012||8 صفحه PDF||سفارش دهید||6154 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Expert Systems with Applications, Volume 39, Issue 5, April 2012, Pages 5784–5791
Currently, tumor markers have been effectively applied for colorectal cancer (CRC) diagnosis. In order to decrease the information loss caused by single cutoff value and improve diagnosis efficiency (DE), we explore the integrative application of multiple tumor markers with multiple cutoff values systematically by developing an optimization algorithm named MVMTM. The effectiveness of the MVMTM is experimentally studied based on a real medical dataset. With MVMTM, the united use of three tumor markers can enhance DE from 0.78 to 0.86. Furthermore, MVMTM has been proved to be better than other baseline machine learning algorithms significantly.
Colorectal cancer (CRC) is one of the most common cancers whose mortality is ranked third in the world. Colonoscopy and fecal occult blood test (FOBT) are the frequently used screening measures for CRC. Although large-scale clinical trials demonstrate that colonoscopy is effective for the diagnosis of CRC, the comparatively high cost and the painful process seriously limit its application (Lewis, 2000). In terms of FOBT, due to the low diagnostic accuracy, it is not an effective method for CRC diagnosis. In this circumstance, a number of tumor markers have been set forth and utilized in CRC diagnosis. For example, in Renji Hospital, a teaching hospital in Shanghai, China, nine different tumor markers, viz. AFP, CEA, CA 19-9, CA 125, CA 153, CA 50, CA 724, CA 211 and CA 242 can be employed for cancer diagnosis. Among them, the most commonly used tumor markers for colorectal cancer diagnosis are CEA, CA 19-9 and CA 50. Usually, single cutoff value is applied to separate the value range of the tumor marker into two segments as normal and abnormal. Some studies demonstrate that diagnostic result can varied widely along with the adjustment of the cutoff value (Körner, Söreide, Stokkeland, & Söreide, 2007). Up to now, many efforts have been dedicated to searching for an appropriate cutoff value so as to achieve the highest diagnosis efficiency (DE) (Armitage et al., 1984, Carriquiry and Pineyro, 1999 and Wichmann et al., 2000). The commonly used approach is to numerate the possible cutoff values and choose the value at which the highest DE can be achieved (Wan and Zhang, 2007 and Weiss et al., 2004). For instance, Körner et al. suggested that the optimal cutoff value for CEA was 4 μg/L (Körner et al., 2007). Carpelan-Holmstrom et al. suggested a cutoff value of 5 μg/L for CEA and 20 U/ml for CA 242 (Carpelan-Holmstrom, Haglund, Kuusela, Jarvinen, & Roberts, 1995). Some researchers found that the combination of different tumor markers can improve DE. Lucarotti et al. stated that the combination use of CEA, CA 19-9, and CA 50 could improve DE significantly in differentiating benign tumor from malignant tumor for the pancreatic cancer (Lucarotti et al., 1991). Moghimi and Ghodosi figured out that the combination use of CEA and CA 19-9 could achieve a higher DE compared with individual usage (Moghimi & Ghodosi, 2007). These findings imply that, instead of being used individually, the different markers should be applied simultaneously and the check results should be considered synthetically. In almost all existing studies, only a single cutoff value is set for each tumor marker (Duffy, 2001, Moertel et al., 1993, Persijn and Hart, 1981 and Wood et al., 1980). In this way, some important information obtained from the diagnosis test will be neglected. Taking CEA as an example, its value can range from 0 to 550 μg/L. If its cutoff value is set to 4 μg/L, the test results of 5 μg/L and 500 μg/L will be simply judged as the same (abnormal) although they are very different from each other. From this point of view, it is too rough to simply separate the broad value range of a tumor marker into two segments of normal and abnormal. Hence, multi-cutoff values should be employed to take more advantages of the test result so as to improve DE. Under this consideration, aiming at improving DE for CRC, a novel diagnosis strategy with multiple tumor markers and multi-cutoff values are studied systematically. An optimization algorithm is designed for setting multi-cutoff values for multiple tumor markers (MVMTM). In this work, three tumor markers, i.e., CEA, CA 19-9, and CA 50, are used simultaneously. And no more than three cutoff values are permitted for each tumor marker. With the algorithm, the optimal cutoff values are calculated out based on a real diagnosis dataset with 124 cases. Furthermore, other 88 cases are employed to validate the effectiveness of the algorithm. This paper is organized as follows. In Section 2, the evaluation method of DE is addressed and the algorithm related technologies including the rough set theory (RST) and the genetic algorithm (GA) are introduced. The details of the MVMTM are elaborated in Section 3. Then, in Section 4, the experimental study is conducted to demonstrate the effectiveness of the algorithm. Furthermore, some discussions are given in Section 5. Section 6 concludes the paper.
نتیجه گیری انگلیسی
In this paper, focusing on how to set the multi-cutoff values for the three tumor markers of CEA, CA 19-9, and CA 50 in CRC diagnosis, an artificial intelligent algorithm entitled MVMTM has been proposed based on the RST and GA. The experimental study demonstrates that, with the MVMTM, the DE of CRC can be enhanced substantially. Although some progress has been made in this study, many issues are still open for future research. For instance, the algorithm should be improved further in terms of the calculation speed and quality. Meanwhile, some other types of combination of different tumor markers should be studied systematically. In addition, the application of this algorithm to diagnosis of other diseases should be explored.