The Effect Of Clinical Decision Support System On The Diagnosis Of Edema Based On Combined Method

Yaghoobi Notash Anaram , Bayat Peyman , Haghighat Shahpar , Yaghoobi Notash Ali

Breast cancer is one of the most common types of cancer among women. The likelihood of a woman suffering from breast cancer during her lifetime is 1.8 and the odds of death by breast cancer is 1.35. Malignant tumors may spread through blood or lymphatic system in other organs. Hence, in order to specify the invasion rate of cancer, it is necessity to excise certain lymph nodes during surgery. When the lymph nodes are removed, protein-rich fluids accumulate in the interstitial tissue, which is called lymphedema. The occurrence of lymphedema depends on different demographic and clinical factors. Getting adequate data from the patient, turning them into categorized information, and making decisions based on such information is a major challenge along the diagnosis and treatment process, and even planning for prevention. Decision support systems are information-based interfaces that have been employed as a solution to this challenge. In this paper, modeling was carried out based on the patients' clinical and demographic factors. The relationship between the underlying variables and the outcome was determined according to the available data. In the implemented model, we have primarily used clustering algorithms such as K-means (K-means clustering aims to partition n observations into k clusters, in which each observation belongs to the cluster with the nearest mean.), support vector machine (SVM), and Linear Support Vector Machine (LSVM). We note that the best results were obtained using the K-means and SVM algorithms. We found 70.94 %, 77.90%, 62.67%, and 75.12% accuracy in the first, second, third, and fourth groups, respectively. K-means algorithm and neural network were further combined with the effective data, whereupon, four different findings were extracted. Next, we combined SVM algorithm and neural network and observed 68.5% accuracy in the first group data. Items with a positive effect on the correct diagnosis were ROM, affected lymph node, excised lymph node, occupation, Herceptin, marital status, paresthesia, education, radio, weight, age, metastases, pain, tumor size, surgery, BMI, and chemotherapy. The final model was designed by combining SVM and neural network with the effective data, resulting in 86.8% accuracy. The final effective data includes: tumor size, BMI, affected lymph node, age, excised lymph node, education, ROM, paresthesia, pain, and marital status.

Volume 12 | Issue 6

Pages: 2204-2247

DOI: 10.5373/JARDCS/V12I6/S20201182