Artificial neural network for prediction of distant metastasis in colorectal cancer

Background and Objectives: Artificial neural networks (ANNs) are flexible and nonlinear models which can be used by clinical oncologists in medical research as decision making tools. This study aimed to predict distant metastasis (DM) of colorectal cancer (CRC) patients using an ANN model. Methods: The data of this study were gathered from 1219 registered CRC patients at the Research Center for Gastroenterology and Liver Disease of Shahid Beheshti University of Medical Sciences, Tehran, Iran (January 2002 and October 2007). For prediction of DM in CRC patients, neural network (NN) and logistic regression (LR) models were used. Then, the concordance index (C index) and the area under receiver operating characteristic curve (AUROC) were used for comparison of neural network and logistic regression models. Data analysis was performed with R 2.14.1 software. Results: The C indices of ANN and LR models for colon cancer data were calculated to be 0.812 and 0.779, respectively. Based on testing dataset, the AUROC for ANN and LR models were 0.82 and 0.77, respectively. This means that the accuracy of ANN prediction was better than for LR prediction. Conclusion: The ANN model is a suitable method for predicting DM and in that case is suggested as a good classifier that usefulness to treatment goals.     

Limits of predictive models using microarray data for breast cancer clinical treatment outcome

Data from microarray studies have been used to develop predictive models for treatment outcome in breast cancer, such as a recently proposed predictive model for antiestrogen response after tamoxifen treatment that was based on the expression ratio of two genes. We attempted to validate this model on an independent cohort of 58 patients with resectable estrogen receptor-positive breast cancer. We measured expression of the genes HOXB13 and IL17BR with real time-quantitative polymerase chain reaction and assessed the association between their expression and outcome by use of univariate logistic regression, area under the receiver-operating-characteristic curve (AUC), a two-sample t test, and a Mann-Whitney test. We also applied standard supervised methods to the original microarray dataset and to another independent dataset from similar patients to estimate the classification accuracy obtainable by using more than two genes in a microarray-based predictive model. We could not validate the performance of the two-gene predictor on our cohort of samples (relation between outcome and the following genes estimated by logistic regression: for HOXB13, odds ratio [OR] = 1.04, 95% confidence interval [CI] = 0.92 to 1.16, P = .54; for IL17BR, OR = 0.69, 95% CI = 0.40 to 1.20, P = .18; and for HOXB13/IL17BR, OR = 1.30, 95% CI = 0.88 to 1.93, P = .18). Similar results were obtained with the AUC, a two-sample two-sided t test, and a Mann-Whitney test. In addition, estimates of classification accuracies applied to two independent microarray datasets highlighted the poor performance of treatment-response predictive models that can be achieved with the sample sizes of patients and informative genes to date.     

Artificial neural networks applied to survival prediction in breast cancer

In this study, we evaluated the accuracy of a neural network in predicting 5-, 10- and 15-year breast-cancer-specific survival. A series of 951 breast cancer patients was divided into a training set of 651 and a validation set of 300 patients. Eight variables were entered as input to the network: tumor size, axillary nodal status, histological type, mitotic count, nuclear pleomorphism, tubule formation, tumor necrosis and age. The area under the ROC curve (AUC) was used as a measure of accuracy of the prediction models in generating survival estimates for the patients in the independent validation set. The AUC values of the neural network models for 5-, 10- and 15-year breast-cancer-specific survival were 0.909, 0.886 and 0.883, respectively.The corresponding AUC values for logistic regression were 0.897, 0.862 and 0.858. Axillary lymph node status (N0 vs. N+) predicted 5-year survival with a specificity of 71% and a sensitivity of 77%. The sensitivity of the neural network model was 91% at this specificity level. The rate of false predictions at 5 years was 82/300 for nodal status and 40/300 for the neural network. When nodal status was excluded from the neural network model, the rate of false predictions increased only to 49/300 (AUC 0. 877). An artificial neural network is very accurate in the 5-, 10- and 15-year breast-cancer-specific survival prediction. The consistently high accuracy over time and the good predictive performance of a network trained without information on nodal status demonstrate that neural networks can be important tools for cancer survival prediction. Copyright Copyright 1999 S. Karger AG, Basel     

Neural networks in the prediction of survival in patients with colorectal cancer

It is important to predict outcome for colorectal cancer patients following surgery, as almost 50% of patients undergoing a potentially curative resection will experience relapse. It is clear that present prognostic categories such as Dukes or TNM staging are too broad, and further refining is required to prognosticate for high-risk subgroups. One approach is to determine a phenotype associated with recurrence. We compared 2 methods of analyzing such data. Pathologic data from a large clinical trial was analyzed for 403 patients. The outcome modeled was disease recurrence. The results from logistic regression analysis and a neural network approach are compared with respect to receiver operator characteristic plots, which estimate the fit of the model. The best logistic regression model gives a result of 66%, and the neural network approach 78%. The conclusion from this study is that the neural network approach is superior to regression analysis. Further analyses are in progress using a larger patient sample size (n > 1000), improved statistical models, and a more refined neural network.     

The development of composite circulating biomarker models for use in anticancer drug clinical development

The development of informative composite circulating biomarkers predicting cancer presence or therapy response is clinically attractive but optimal approaches to modeling are as yet unclear. This study investigated multidimensional relationships within an example panel of serum insulin‐like growth factor (IGF) peptides using logistic regression (LR), fractional polynomial (FP), regression, artificial neural networks (ANNs) and support vector machines (SVMs) to derive predictive models for colorectal cancer (CRC). Two phase 2 biomarker validation analyses were performed: controls were ambulant adults (n = 722); cases were: (i) CRC patients (n = 100) and (ii) patients with acromegaly (n = 52), the latter as “positive” discriminators. Serum IGF‐I, IGF‐II, IGF binding protein (IGFBP)‐2 and ‐3 were measured. Discriminatory characteristics were compared within and between models. For the LR, FP and ANN models, and to a lesser extent SVMs, the addition of covariates at several steps improved discrimination characteristics. The optimum biomarker combination discriminating CRC vs. controls was achieved using ANN models [sensitivity, 94%; specificity, 90%; accuracy, 0.975 (95% CIs: 0.948 1.000)]. ANN modeling significantly outperformed LR, FP and SVM in terms of discrimination (p < 0.0001) and calibration. The acromegaly analysis demonstrated expected high performance characteristics in the ANN model [accuracy, 0.993 (95% CIs: 0.977, 1.000)]. Curved decision surfaces generated from the ANNs revealed the potential clinical utility. This example demonstrated improved discriminatory characteristics within the composite biomarker ANN model and a final model that outperformed the three other models. This modeling approach forms the basis to evaluate composite biomarkers as pharmacological and predictive biomarkers in future clinical trials.     

Combining artificial neural networks and transrectal ultrasound in the diagnosis of prostate cancer

Arguably the most important step in the prognosis of prostate cancer is early diagnosis. More than 1 million transrectal ultrasound (TRUS)-guided prostate needle biopsies are performed annually in the United States, resulting in the detection of 200,000 new cases per year. Unfortunately, the urologist's ability to diagnose prostate cancer has not kept pace with therapeutic advances; currently, many men are facing the need for prostate biopsy with the likelihood that the result will be inconclusive. This paper will focus on the tools available to assist the clinician in predicting the outcome of the prostate needle biopsy. We will examine the use of "machine learning" models (artificial intelligence), in the form of artificial neural networks (ANNs), to predict prostate biopsy outcomes using prebiopsy variables. Currently, six validated predictive models are available. Of these, five are machine learning models, and one is based on logistic regression. The role of ANNs in providing valuable predictive models to be used in conjunction with TRUS appears promising. In the few studies that have compared machine learning to traditional statistical methods, ANN and logistic regression appear to function equivalently when predicting biopsy outcome. With the introduction of more complex prebiopsy variables, ANNs are in a commanding position for use in predictive models. Easy and immediate physician access to these models will be imperative if their full potential is to be realized.     

Predictive model for 5-year mortality after breast cancer surgery in Taiwan residents

Background: Few studies of breast cancer surgery outcomes have used longitudinal data for more than 2 years. This study aimed to validate the use of the artificial neural network (ANN) model to predict the 5-year mortality of breast cancer patients after surgery and compare predictive accuracy between the ANN model, multiple logistic regression(MLR) model, and Cox regression model.Methods: This study compared the MLR, Cox, and ANN models based on clinical data of 3632 breast cancer patients who underwent surgery between 1996 and 2010. An estimation dataset was used to train the model, and a validation dataset was used to evaluate model performance. The sensitivity analysis was also used to assess the relative significance of input variables in the prediction model.Results: The ANN model significantly outperformed the MLR and Cox models in predicting 5-year mortality, with higher overall performance indices. The results indicated that the 5-year postoperative mortality of breast cancer patients was significantly associated with age, Charlson comorbidity index (CCI), chemotherapy, radiotherapy, hormone therapy, and breast cancer surgery volumes of hospital and surgeon (all P < 0.05). Breast cancer surgery volume of surgeon was the most influential (sensitive) variable affecting 5-year mortality, followed by breast cancer surgery volume of hospital, age, and CCI.Conclusions: Compared with the conventional MLR and Cox models, the ANN model was more accurate in predicting 5-year mortality of breast cancer patients who underwent surgery. The mortality predictors identified in this study can also be used to educate candidates for breast cancer surgery with respect to the course of recovery and health outcomes.     

Omission of histologic grading from clinical decision making may result in overuse of adjuvant therapies in breast cancer: results from a nationwide study.

PURPOSE: To investigate the influence of routinely performed histologic grading on breast cancer outcome prediction and patient selection for adjuvant therapy. PATIENTS AND METHODS: The analysis is based on a cohort of 2,842 women diagnosed with breast cancer and comprising 91% of all breast cancers diagnosed in five defined geographical regions in Finland in 1991 through 1992. Data on clinicopathologic factors and follow-up were collected from hospital case records and national registries. Histologic grade assessed at diagnosis and other clinicopathologic data were available for 1,554 operable unilateral invasive carcinomas. The relative value of grade with respect to competing prognostic factors was estimated with the Cox proportional hazards model and logistic regression. Interactions and nonlinearity of factors were accounted for by using an artificial neural network. RESULTS: Histologic grade was correlated strongly with survival in the entire series and in all subgroups studied. Women with well-differentiated node-negative cancer had a 97% 5-year distant disease-free survival rate as compared with 78% for women with poorly differentiated cancer. Grade was an independent prognostic factor in multivariate models and increased the predictive accuracy of a neural network model. Inclusion of grade data in a Cox multivariate model based on tumor size and hormone receptor status in node-negative cancer increased the proportion of patients with 5% or less risk for distant recurrence at 5 years from 15% to 54%. CONCLUSION: Even when assessed by pathologists who have no special training in breast cancer pathology, histologic grade has substantial and independent prognostic value in breast cancer. Omission of grading from clinical decision making may result in considerable overuse of adjuvant therapies.     

预测上皮性卵巢癌复发机器学习模型及列线图的建立

目的:本研究拟基于机器学习及Cox回归开发上皮性卵巢癌复发机器学习模型及列线图。方法:回顾性分析2010年01月至2020年12月于云南省肿瘤医院确诊739例Ⅲ-Ⅳ期EOC患者的医疗记录。收集患者的基本信息、手术、化疗细节和预后结果。使用单多因素逻辑回归及Cox回归筛选变量,使用5种机器学习算法基于单多因素逻辑回归的结果构建预测模型,采用10折交叉验证方法评估模型性能。基于Cox回归结果开发列线图。结果:739例患者中,399（54.0%）例最终发生了复发,340（46.0%）例未复发。复发患者分期以ⅢC期为主,占59.1%,病理类型以浆液性癌为主,占91.0%。单多因素逻辑回归显示围手术期化疗周期、术后残余病灶、手术方式、新辅助化疗是与复发独立相关的4个变量,基于这些变量和FIGO分期建立5个机器学习模型,XGBoost在识别复发病例方面表现最佳,AUC为0.775。Cox回归分析显示,术前局部灌注化疗、残余病灶直径、围手术期化疗周期、手术方式是影响复发的独立危险因素,基于上述因素开发了晚期上皮性卵巢癌患者复发的预测列线图。结论:机器学习模型和列线图可早期识别卵巢癌复发,通过早期识别可改善晚期卵巢癌预后。     

The use of artificial neural network analysis to improve the predictive accuracy of prostate biopsy in the Japanese population

OBJECTIVE: We examined the efficacy of an artificial neural network analysis (ANNA) based on parameters available from previously existing examinations for improving the predictive accuracy of prostate cancer screening in the Japanese population. METHODS: Two hundred and twenty-eight patients with prostate-specific antigen (PSA) of 2-10 ng/ml were enrolled in this study. Two artificial neural network analysis (ANNA) models were constructed: ANNA1 with patient age, total PSA, free to total PSA ratio, prostate volume, transition zone volume (TZ), PSA density (PSAD) and PSA-TZ density (PSATZ) as input variables, and ANNA2 with presumed circle area ratio (PCAR), digital rectal examination (DRE) findings and chief complaint added as variables. The predictive accuracies of the ANNA models were compared with conventional PSA and volume-related parameters and a logistic regression (LR) model by receiver operating characteristic (ROC) curve analysis. RESULTS: Of 228 patients, 58 (25.5%) were diagnosed with prostate cancer. While ANNA2 had a slightly larger area under the curve (AUC) than ANNA1 (0.782 versus 0.793, P = 0.8477), the AUC of ANNA2 was significantly greater than those of ln(PSA), PSAD, PSATZ and free to total PSA ratio (P = 0.0004, 0.0230, 0.0304, and 0.0037, respectively). The accuracy of ANNA2 was significantly better than that of LR analysis at 90 and 95% sensitivity levels (P = 0.0051 and P < 0.0001, respectively). At 95% sensitivity level, ANNA2 reduced unnecessary biopsies by 40.0% with a negative predictive value of 95.7%. CONCLUSIONS: To determine the indication of prostate biopsy for PSA value in the range of 2-10 ng/ml, the ANNA model has the possibility to reduce unnecessary biopsies without missing many cases of cancers.     

Comparison of nomograms with other methods for predicting outcomes in prostate cancer: a critical analysis of the literature

PURPOSE: Accurate estimates of risk are essential for physicians if they are to recommend a specific management to patients with prostate cancer. Accurate risk estimates are also required for clinical trial design, to ensure homogeneous patient groups. Because there is more than one model available for prediction of most outcomes, model comparisons are necessary for selection of the best model. We describe the criteria based on which to judge predictive tools, describe the limitations of current predictive tools, and compare the different predictive methodologies that have been used in the prostate cancer literature. EXPERIMENTAL DESIGN: Using MEDLINE, a literature search was done on prostate cancer decision aids from January 1966 to July 2007. RESULTS: The decision aids consist of nomograms, risk groupings, artificial neural networks, probability tables, and classification and regression tree analyses. The following considerations need to be applied when the qualities of predictive models are assessed: predictive accuracy (internal or ideally external validation), calibration (i.e., performance according to risk level or in specific patient subgroups), generalizability (reproducibility and transportability), and level of complexity relative to established models, to assess whether the new model offers advantages relative to available alternatives. Studies comparing decision aids have shown that nomograms outperform the other methodologies. CONCLUSIONS: Nomograms provide superior individualized disease-related risk estimations that facilitate management-related decisions. Of currently available prediction tools, the nomograms have the highest accuracy and the best discriminating characteristics for predicting outcomes in prostate cancer patients.     

Machine learning approaches for predicting 5-year breast cancer survival: A multicenter study

The study used clinical data to develop a prediction model for breast cancer survival. Breast cancer prognostic factors were explored using machine learning techniques. We conducted a retrospective study using data from the Taipei Medical University Clinical Research Database, which contains electronic medical records from three affiliated hospitals in Taiwan. The study included female patients aged over 20 years who were diagnosed with primary breast cancer and had medical records in hospitals between January 1, 2009 and December 31, 2020. The data were divided into training and external testing datasets. Nine different machine learning algorithms were applied to develop the models. The performances of the algorithms were measured using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1-score. A total of 3914 patients were included in the study. The highest AUC of 0.95 was observed with the artificial neural network model (accuracy, 0.90; sensitivity, 0.71; specificity, 0.73; PPV, 0.28; NPV, 0.94; and F1-score, 0.37). Other models showed relatively high AUC, ranging from 0.75 to 0.83. According to the optimal model results, cancer stage, tumor size, diagnosis age, surgery, and body mass index were the most critical factors for predicting breast cancer survival. The study successfully established accurate 5-year survival predictive models for breast cancer. Furthermore, the study found key factors that could affect breast cancer survival in Taiwanese women. Its results might be used as a reference for the clinical practice of breast cancer treatment.     

Artificial neural network analysis for predicting pathological stage of clinically localized prostate cancer in the Japanese population

BACKGROUND: Although prostate cancer has been prevalent in Japan, there has been no particular model for predicting the pathological stage in the Japanese population. We examined whether artificial neural network analysis (ANNA), which is a relatively new diagnostic tool in prostate cancer, can be one of the predictive methods for predicting organ confinement, compared with the traditional logistic regression model, in the Japanese population for the first time. METHODS: The study population comprised 178 men who underwent radical prostatectomy at our institutions between October 1992 and May 1999. As additional pretreatment parameters to the preoperative serum PSA level, clinical TNM classification and biopsy Gleason score, the percentage of number of cores exhibiting traces of tumor, maximum tumor length in biopsy cores, PSA density and patient age were used. The predictive ability of ANNA with several parameters for a set of 36 randomly selected test data was compared with those of logistic regression analysis and 'Partin Tables' by area under the receiver operating characteristics (ROC) curve analysis. RESULTS: Of 178 patients, 97 (54.5%) had organ-confined disease but 81 (45.5%) had locally advanced disease. With three parameters, the area under the ROC curve of ANNA (0.825 +/- 0.071) was larger than those for logistic regression (0.782 +/- 0.079) and Partin Tables (0.756 +/- 0.087), but not to a significant extent (P = 0.690 and 0.541). Although the expansion of the parameters did not increase the difference in area under the ROC curve between the best ANNA and logistic regression (0.899 +/- 0.053 and 0.873 +/- 0.065, respectively), the difference between the best ANNA and Partin Tables did not reach but approached statistical significance (P = 0.157). CONCLUSION: Although more modeling optimization is necessary to improve the predictive accuracy and generalizability of ANNA, we suggest that there is the possibility for this new predictive method to evolve in the analysis of clinical staging of prostate cancer.     

Evaluation of predictive models in daily practice for the identification of patients with Lynch syndrome

The optimal strategy for identifying patients with Lynch syndrome among patients with newly diagnosed colorectal cancer (CRC) is still debated. Several predictive models (e.g., MMRpredict, PREMM1,2 and MMRpro) combining personal and familial data have recently been developed to quantify the risk that a given patient with CRC carries a Lynch syndrome-causing mutation. Their clinical applicability to patients with CRC from the general population requires evaluation. We studied a consecutive series of 214 patients with newly diagnosed CRC characterized for tumor microsatellite instability (MSI), somatic BRAF mutation, MLH1 promoter methylation and mismatch repair (MMR) gene germline mutation status. The performances of the models for identifying MMR mutation carriers (8/214, 3.7%) were evaluated and compared to the revised Bethesda guidelines and a molecular strategy based on MSI testing in all patients followed by the exclusion of MSI-positive sporadic cases from mutational testing by screening for BRAF mutation and MLH1 promoter methylation. The sensitivities of the three models, at the lowest thresholds proposed, were identical (75%), with similar numbers of probands eligible for further MSI testing (almost half the patients). In our dataset, the prediction models gave no better discrimination than the revised Bethesda guidelines. Both approaches failed to identify two of the eight mutation carriers (the same two patients, aged 67 and 81 years, both with no family history). Thus, like the revised Bethesda guidelines, predictive models did not identify all patients with Lynch syndrome in our series of consecutive CRC. Our results support systematic screening for MMR deficiency in all new CRC cases.     

Artificial Intelligence System for Predicting Prostate Cancer Lesions from Shear Wave Elastography Measurements

(1) Objective: To design an artificial intelligence system for prostate cancer prediction using the data obtained by shear wave elastography of the prostate, by comparing it with the histopathological exam of the prostate biopsy specimens. (2) Material and methods: We have conducted a prospective study on 356 patients undergoing transrectal ultrasound-guided prostate biopsy, for suspicion of prostate cancer. All patients were examined using bi-dimensional shear wave ultrasonography, which was followed by standard systematic transrectal prostate biopsy. The mean elasticity of each of the twelve systematic biopsy target zones was recorded and compared with the pathological examination results in all patients. The final dataset has included data from 223 patients with confirmed prostate cancer. Three machine learning classification algorithms (logistic regression, a decision tree classifier and a dense neural network) were implemented and their performance in predicting the positive lesions from the elastographic data measurements was assessed. (3) Results: The area under the curve (AUC) results were as follows: for logistic regression—0.88, for decision tree classifier—0.78 and for the dense neural network—0.94. Further use of an upsampling strategy for the training set of the neural network slightly improved its performance. Using an ensemble learning model, which combined the three machine learning models, we have obtained a final accuracy of 98%. (4) Conclusions: Bi-dimensional shear wave elastography could be very useful in predicting prostate cancer lesions, especially when it benefits from the computational power of artificial intelligence and machine learning algorithms.     

Variation in the use of chemotherapy in lung cancer

Factors influencing the use of chemotherapy for the initial (6 months) treatment of lung cancer in South East England were investigated. The variables explored as possibly influencing the use of chemotherapy were sex, age, the year of diagnosis, the type of lung cancer, the stage, the index of multiple deprivation and the cancer network of residence. Chi2 analysis and multivariate logistic regression models were used to examine the effect of each of the variables on the use of chemotherapy. The results showed a highly significant trend in use of chemotherapy over time; the adjusted proportion of patients receiving chemotherapy increasing from 13.6% in 1994 to 29.3% in 2003. However, age, cancer network and type of lung cancer had the strongest influence on the use of chemotherapy. This finding is important when we consider that the NHS Cancer Plan aims at improving inequalities in cancer care in the UK.     

Multivariate analysis of prognostic factors in metastatic breast cancer

Univariate and multivariate analyses were conducted on data collected from the records of 619 patients with metastatic breast cancer in whom an Adriamycin-containing chemotherapeutic regimen was used. Using a forward, stepwise logistic regression procedure, several models or equations in which a small number of pretreatment factors were incorporated were generated and the probability of response to therapy was accurately predicted. The predictive ability of these models was tested retrospectively in 546 of the 619 patients from whom the data were derived and prospectively in a new population of 200 patients with metastatic breast cancer also treated with a therapeutically equivalent Adriamycin combination. Using similar univariate techniques, pretreatment factors were correlated with the length of survival after therapy. The proportional hazard model of Cox was used to develop a regression model relating survival to pretreatment characteristics in much the same manner as that of the response model. The total population of the initial group of patients was divided according to four levels of hazard ratio, and survival distributions were compared. This model also was tested progressively and its predictive capability was confirmed. The prediction of individual outcome is a valuable capability in the comparison of clinical trials and the continuing evaluation of biologic changes in patients with metastatic carcinoma; such a method is described in this paper.     

Nuclear morphometry predicts disease-free interval for clinically localized adenocarcinoma of the prostate treated with definitive radiation therapy

Men treated for prostate cancer often have unexpected outcomes despite predictive models based on stage, grade and prostate-specific antigen (PSA). Previous results have indicated that nuclear morphometry can predict patient outcome in urologic malignancies. Application of this analytical method in prostate cancer treated with radiation therapy is limited. We have evaluated the predictive ability of nuclear morphometry in such patients. Histologic sections from 23 men with clinically localized adenocarcinoma of the prostate treated with radiation therapy were studied. Nuclear morphometric parameters were assessed using a previously described and validated system. Univariate and multivariate logistic regression analyses and a Cox proportional hazards model were used to assess the ability of nuclear morphometric parameters to predict recurrence and disease-free interval. Ten patients had no recurrence with median follow-up of 47. 5 months, while 13 had recurrence. Gleason grade was not predictive of treatment outcome. Pre-treatment PSA data, available for only 11 patients, were predictive of treatment outcome. Several nuclear morphometric parameters predicted recurrence, including upper quartile of suboptimal circle fit and upper quartile of feret-diameter ratio. A prognostic factor score incorporating these 2 parameters was derived, which predicted disease-free interval (p = 0.0014). Int. J. Cancer (Pred. Oncol.) 84:594-597, 1999.     

Prognostic models in patients with non-small-cell lung cancer using artificial neural networks in comparison with logistic regression

It is difficult to precisely predict the outcome of each individual patient with non-small-cell lung cancer (NSCLC) by using conventional statistical methods and ordinary clinico-pathological variables. We applied artificial neural networks (ANN) for this purpose. We constructed a prognostic model for 125 NSCLC patients with 17 potential input variables, including 12 clinico-pathological variables (age, sex, smoking index, tumor size, p factor, pT, pN, stage, histology) and 5 immunohistochemical variables (p27 percentage, p27 intensity, p53, cyclin D1, retinoblastoma (RB)), by using the parameter-increasing method (PIM). Using the resultant ANN model, prediction was possible in 104 of 125 patients (83%, judgment ratio ( JR )) and accuracy for prediction of survival at 5 years was 87%. On the other hand, JR and survival prediction accuracy in the logistic regression (LR) model were 37% and 78%, respectively. In addition, ANN outperformed LR for prediction of survival at 1 or 3 years. In these cases, PIM selected p27 intensity and cyclin D1 for the 3-year survival model and p53 for the 1-year survival model in addition to clinico-pathological variables. Finally, even in an independent validation data set of 48 patients, who underwent surgery 10 years later, the present ANN model could predict outcome of patients at 5 years with the JR and accuracy of 81% and 77%, respectively. This study demonstrates that ANN is a potentially more useful tool than conventional statistical methods for predicting survival of patients with NSCLC and that inclusion of relevant molecular markers as input variables enhances its predictive ability. (Cancer Sci 2003; 94: 473–477)