| PurposeCervical cancer is the fourth most common cancer in women worldwide,with high morbidity and mortality.The prediction of cervical cancer recurrence and death after formal treatment remains a difficult problem.At present,the prediction model of postoperative survival of cervical cancer has not been recommended by consensus and guidelines.Probabilistic Neural Networks(PNN)can realize the functions of nonlinear and linear learning algorithms.and are widely used in pattern classification problems and have been used in many aspects of medical tumor prediction.In order to achieve the goal of predicting individual long-term cancer survival to guide treatment decisions,the ability to survive at 10 years in cervical cancer patients undergoing surgical treatment was analyzed using the PNN prediction model.MethodsA total of 245 patients with cervical cancer in our hospital from January 2012 to December 2012,The input data set came from 126 patients treated with radical hysterectomy plus pelvic lymph node resection plus aortic lymph node resection from the International Federation of Gynecology and Obstetrics(FIGO)(2018 edition)in patients with stage ⅠA2-ⅡB cervical cancer.Four demographic parameters,12 clinicpathological features,and six selected perioperative variables were identified as recognized prognostic factors for cervical cancer for each patient,and computer simulations were performed using DTREG software.the predictive power of the model was determined based on its error,sensitivity and specificity,as well as the Area under the receiver operating characteristic curve(AUROC).The results of the PNN prediction model were compared with the results of Logistic Regression(LR)and Single Decision Tree(SDT)as reference models,and the current survival status of all patients was analyzed.Results1、The 126 patients in this study ranged in age from 36.23 to 70.06 years with a median age of 51.82 years.Of the 120 patients,35 had other conditions(hypertension 20,diabetes 4.ischemic heart disease 7,and other conditions 4),and 50 were postmenopausal.2、Clinical characteristics of patients showed that 20 patients were in stage IA2.57 patients were in stage ⅠB1,6 patients were in stage ⅠB2,10 patients were in stage ⅡA,and 33 patients were in stage ⅡB.113 patients presented with squamous cell carcinoma.and the remaining 13 had other histological types.34 patients showed highly differentiated cervical cancer tissues,71 patients showed moderate differentiation,and 21 patients showed low differentiation.Cervical cancer tumors<4 cm were found in 88 patients,and ≥4 cm were found in the remaining 38 patients.Ninety-three patients had negative positive lymph nodes and the remaining 33 were positive.Among the 33 patients with positive lymph nodes,24 had lymph node rate ≤50.00%and 9 had lymph node rate>50.00%.During follow-up,six patients were lost to follow-up,and at the last follow-up.92 patients survived and 28 patients died.The overall 10-year survival rate was 76.67%.3、Using the PNN model.we predicted the 10-year overall survival rate of patients with cervical cancer who underwent radical hysterectomy+pelvic lymph node dissection ± abdominal aortic lymph node dissection surgery.Out of the 92 surviving patients,only 11 were misclassified as false positive,and out of the 28 deceased cases.14 were misclassified as false negative.The sensitivity and specificity of the PNN model were 0.933 and 0.708,respectively.The error rate was 2.50%,and the AUROC was 0.866.We also evaluated the performance of the LR and SDT models.The sensitivity of the LR model was 0.958,specificity was 0.575,error rate was 3.30%.and the AUROC was 0.742.The sensitivity of the SDT model was 0.967,specificity was 0.575,error rate was 3.20%,and the AUROC was 0.765.Comparing the three models,we found that the sensitivity of the PNN model was slightly lower than that of the LR and SDT models,although this difference was not statistically significant(p>0.05).However,the error rate of the PNN model was significantly lower than that of the LR and SDT models(p<0.05),and the specificity was significantly higher(p<0.05).Additionally,the AUROC of the PNN model was greater than that of the LR and SDT models(p<0.05).4、PNN model predicted that the overall survival rate of cervical cancer patients 10 years after surgery was 67.57%.LR model predicted that the overall survival rate of cervical cancer patients 10 years after surgery was 53.29%.SDT model predicted that the overall survival rate of cervical cancer patients 10 years after surgery was 57.14%.and the actual survival rate of cervical cancer patients 10 years after surgery was 76.67%.Conclusion1、The PNN model outperforms both the LR and SDT models in terms of specificity,error rate,and AUROC.2、While the PNN model exhibits high specificity and a low error rate.its sensitivity may be comparatively lower due to an imbalance in the survival/death data resulting from a limited sample size or a scarcity of death data.3、The simplicity and swiftness of PNN’s training process represent the model’s most crucial operational advantages.Its training time is only marginally greater than the time taken to read in the data.Additionally.PNN can handle faulty samples with remarkable tolerance and guarantees the optimal solution as per the Bayesian criterion.provided there is sufficient training information.irrespective of the classification problem’s complexity.4、PNN stands out from many artificial neural networks since it operates in total parallelism,delivering unique benefits in predicting and diagnosing cervical cancer cases.The model can reliably and effectively forecast cervical cancer patients’ 10-year survival rates after undergoing surgery.making it a dependable tool in the cancer treatment decision-making process. |
PDF Full Text Request