Optimization Research On Cancer Decision For Diagnosis/Treatment Using Data Mining

The development of health care is the main factor to improve the life quality and average life years of human beings. Due to the worse global environment and the lag of medical techonology development, cancer has become one of the biggest killers of human. Health care is the application field for many modern scientific techonologies. As one of the hot topicts, the interdisciplinary collaboration between medicine and engineering (mathematics) has received great attention. Benefiting from the close collaboration with physicians, we are successful to find out the interdisciplinary collaboration point-research on cancer diagnosis and treatment decision optimization.Considering the complexity and diversity of cancers, this paper chooses the colorectal cancer for deep study on the optimization problem of decisions on key points of medical process. The data source of data mining in this paper is the actual clinical data from hospitals and data extracted from medical papers. The Chinese and American physicians provide helpful assistance to this research regarding medical knowledge which guarantees the reliability and practicality of this paper. The curement of cancer is inevitably related to the right diagnosis and treatment process. Therefore, the main content of this research covers both diagnosis and treatment process for colorectal cancer. Firstly, this research studies the multi-cutoff points for multiple tumor markers in the diagnosis. It describes the current status of tumor markers'application and points out the problems which requires the research on multi-cutoff points for multiple tumor markers for the purpose of decreasing the information loss and improving the diagnosis accuracy. To this end, we design a specific algorithm for this research which is applied to the practical data obtained from a teaching hospital in Shanghai. The experiment outcome shows that this algorithm is successful to enhance the diagnosis accuracy from 78% to 87% compared to the clinical serial test. Futhermore, it is also significantly better than other algorithms including SVM, BPNN, KNN and Decision Tree.Secondly, this research also discusses the optimization for serum tumer markers'combination. Presently, the medical experts have achieved a consensus that the combination usage of multiple tumor markers is effective to improve the diagnosis accuracy. However, it is still an unresolved problem that which combination is the best. In this scenario, there exists a wide variability in the type and number of routinely used markers so that, sometimes, patients may receive redundant or insufficient checks. This research tries to design an algorithm to complete the mutli-cuttoff setting and combination optimiazation synchronously. The clinical experiments show that the best combinationa among five tumor makers (AFP, CEA, CA199, CA50 and CA125) is CEA+CA50. This conclusion helps us to decrease the total diansis cost of rumor markers by 56%. Furthermore, the diagnosis accuracy is improved from 67% to 74%.Thirdly, this research optimizes models on cancer chemotherapy treatment planning. The killing effect of chemotherapy drugs on both cancerous cells and normal cells, and the drug ressitance happening during the treatment process are the three factors have to be considered in the optimaization model for chemotherapy. This research introduces the objective function and subjection fuctions of the optimal control model and summarizes four weaknesses of current research. Therefore, this research puts forward a Markov Decision Process (MDP) model for chemotherapy in order to resolve or relieve those four problems. It is the first MDP model for such research. This section applies a new parameter Dt to represent the drug ressitance and sets the Pt (? s,a) to the MDP model. A case study is calculated with the new model and several suggestions are achieved.Finally, this research also makes the cost-effectiveness anlaysis of adjuvant chemotherapy FOLFOX for patients with stage II colon cancer. Because adjuvant chemotherapy has been included in the practice guideline for stage III colon cancer but not for stage II colon cancer, it is necessary to have a qualitative research on the feasibility of adjuvant chemotherapy for stage II colon cancer patients. Based on the logical, precise and accurate research attitude of engineering, this section builds two Markov models for both chemotherapy period and follow-up period. Most of the data are collected from clinical papers and some are calculated with calibration algorithm. The result demonstrates that adjuvant chemotherapy with FOLFOX is cost-effective for patients with stage II colon cancer who are younger than or equal to 65.This research makes a great contribution to the medical research to enable it to transfer from qualitative research to quantitative research. Some research outcomes have been approved by pracital clinical diagnosis and treatment. Apart from improving the quality of health care service, it also has great value from economic and societal perspectives. The research is not limited to the colorectal cancer and it also provides useful methdologies and suggestions to other cancers diagnosis and treatment. To summarize, this research is meaningful for academic research and practical implications.