Use of artificial neural networks to predict biological outcomes for patients receiving radical radiotherapy of the prostate

Outcome assessment of patients with metastatic renal cell carcinoma under systemic therapy using artificial neural networks

BACKGROUND

The outcome of patients with advanced renal cell carcinoma (RCC) under systemic therapy shows remarkable variability, and there is a need to identify prognostic parameters that allow individual prognostic stratification and selection of optimal therapy. Artificial neural networks (ANN) are software systems that can be trained to recognize complex data patterns. In this study, we used ANNs to identify poor prognosis of patients with RCC based on common clinical parameters available at the beginning of systemic therapy.

PATIENTS AND METHODS

Data from patients with RCC who started systemic therapy were collected prospectively in a single center database; 175 data sets with follow-up data (median, 36 months) were available for analysis. Age, sex, body mass index, performance status, histopathologic parameters, time interval between primary tumor and detection of metastases, type of systemic therapy, number of metastases, and metastatic sites were used as input data for the ANN. The target variable was overall survival after 36 months. Logistic regression models were constructed by using the same variables.

RESULTS

Death after 36 months occurred in 26% of the patients in the tyrosine kinase inhibitors group and in 37% of the patients in the immunotherapy group (P = .22). ANN achieved 95% overall accuracy and significantly outperformed logistic regression models (78% accuracy). Pathologic T classification, invasion of vessels, and tumor grade had the highest impact on the network's decision.

CONCLUSION

ANN is a promising approach for individual risk stratification of patients with advanced RCC under systemic therapy, based on clinical parameters, and can help to optimize the therapeutic strategy.

Use of machine learning methods for prediction of acute toxicity in organs at risk following prostate radiotherapy

PURPOSE

The goal of this study is to investigate the advantages of large scale optimization methods vs conventional classification techniques in predicting acute toxicity for urinary bladder and rectum due to prostate irradiation.

METHODS

Clinical and dosimetric data of 321 patients undergoing prostate conformal radiotherapy were recorded. Gastro-intestinal and genito-urinary acute toxicities were scored according to the Radiation Therapy Oncology Group/European Organization for Research and Treatment of Cancer (RTOG/EORTC) scale. Patients were classified in two categories to separate mild (Grade < 2) from severe toxicity levels (Grade > 2). Machine learning methods at different complexity were implemented to predict toxicity as a function of multiple variables. The first approach consisted of a large scale optimization method, based on genetic algorithms (GAs) and artificial neural networks (ANN). The second approach was a binary classifier based on support vector machines (SVM).

RESULTS

The ANN and SVM-based solutions showed comparable prediction accuracy, exhibiting an area under the receiver operating characteristic (ROC) curve of 0.7. Different sensitivity and specificity features were measured for the two approaches. The ANN algorithm showed enhanced sensitivity if combined with appropriate classification criteria.

CONCLUSIONS

The results demonstrate that high sensitivity in toxicity prediction can be achieved with optimized ANNs, that are put forward to represent a valuable support in medical decisions. Future studies will be focused on enlarging the available patient database to increase the reliability of toxicity prediction algorithms and to define optimal classification criteria.

Early prediction of response to radiotherapy and androgen-deprivation therapy in prostate cancer by repeated functional MRI: a preclinical study

Background

In modern cancer medicine, morphological magnetic resonance imaging (MRI) is routinely used in diagnostics, treatment planning and assessment of therapeutic efficacy. During the past decade, functional imaging techniques like diffusion-weighted (DW) MRI and dynamic contrast-enhanced (DCE) MRI have increasingly been included into imaging protocols, allowing extraction of intratumoral information of underlying vascular, molecular and physiological mechanisms, not available in morphological images. Separately, pre-treatment and early changes in functional parameters obtained from DWMRI and DCEMRI have shown potential in predicting therapy response. We hypothesized that the combination of several functional parameters increased the predictive power.

Methods

We challenged this hypothesis by using an artificial neural network (ANN) approach, exploiting nonlinear relationships between individual variables, which is particularly suitable in treatment response prediction involving complex cancer data. A clinical scenario was elicited by using 32 mice with human prostate carcinoma xenografts receiving combinations of androgen-deprivation therapy and/or radiotherapy. Pre-radiation and on days 1 and 9 following radiation three repeated DWMRI and DCEMRI acquisitions enabled derivation of the apparent diffusion coefficient (ADC) and the vascular biomarker K^trans, which together with tumor volumes and the established biomarker prostate-specific antigen (PSA), were used as inputs to a back propagation neural network, independently and combined, in order to explore their feasibility of predicting individual treatment response measured as 30 days post-RT tumor volumes.

Results

ADC, volumes and PSA as inputs to the model revealed a correlation coefficient of 0.54 (p < 0.001) between predicted and measured treatment response, while K^trans, volumes and PSA gave a correlation coefficient of 0.66 (p < 0.001). The combination of all parameters (ADC, K^trans, volumes, PSA) successfully predicted treatment response with a correlation coefficient of 0.85 (p < 0.001).

Conclusions

We have in a preclinical investigation showed that the combination of early changes in several functional MRI parameters provides additional information about therapy response. If such an approach could be clinically validated, it may become a tool to help identifying non-responding patients early in treatment, allowing these patients to be considered for alternative treatment strategies, and, thus, providing a contribution to the development of individualized cancer therapy.

Modeling late rectal toxicities based on a parameterized representation of the 3D dose distribution

Many models exist for predicting toxicities based on dose-volume histograms (DVHs) or dose-surface histograms (DSHs). This approach has several drawbacks as firstly the reduction of the dose distribution to a histogram results in the loss of spatial information and secondly the bins of the histograms are highly correlated with each other. Furthermore, some of the complex nonlinear models proposed in the past lack a direct physical interpretation and the ability to predict probabilities rather than binary outcomes. We propose a parameterized representation of the 3D distribution of the dose to the rectal wall which explicitly includes geometrical information in the form of the eccentricity of the dose distribution as well as its lateral and longitudinal extent. We use a nonlinear kernel-based probabilistic model to predict late rectal toxicity based on the parameterized dose distribution and assessed its predictive power using data from the MRC RT01 trial (ISCTRN 47772397). The endpoints under consideration were rectal bleeding, loose stools, and a global toxicity score. We extract simple rules identifying 3D dose patterns related to a specifically low risk of complication. Normal tissue complication probability (NTCP) models based on parameterized representations of geometrical and volumetric measures resulted in areas under the curve (AUCs) of 0.66, 0.63 and 0.67 for predicting rectal bleeding, loose stools and global toxicity, respectively. In comparison, NTCP models based on standard DVHs performed worse and resulted in AUCs of 0.59 for all three endpoints. In conclusion, we have presented low-dimensional, interpretable and nonlinear NTCP models based on the parameterized representation of the dose to the rectal wall. These models had a higher predictive power than models based on standard DVHs and their low dimensionality allowed for the identification of 3D dose patterns related to a low risk of complication.

Quantitative Analyses of Normal Tissue Effects in the Clinic (QUANTEC): an introduction to the scientific issues

Advances in dose-volume/outcome (or normal tissue complication probability, NTCP) modeling since the seminal Emami paper from 1991 are reviewed. There has been some progress with an increasing number of studies on large patient samples with three-dimensional dosimetry. Nevertheless, NTCP models are not ideal. Issues related to the grading of side effects, selection of appropriate statistical methods, testing of internal and external model validity, and quantification of predictive power and statistical uncertainty, all limit the usefulness of much of the published literature. Synthesis (meta-analysis) of data from multiple studies is often impossible because of suboptimal primary analysis, insufficient reporting and variations in the models and predictors analyzed. Clinical limitations to the current knowledge base include the need for more data on the effect of patient-related cofactors, interactions between dose distribution and cytotoxic or molecular targeted agents, and the effect of dose fractions and overall treatment time in relation to nonuniform dose distributions. Research priorities for the next 5-10 years are proposed.

Assessing correlations between the spatial distribution of the dose to the rectal wall and late rectal toxicity after prostate radiotherapy: an analysis of data from the MRC RT01 trial (ISRCTN 47772397)

Many studies have been performed to assess correlations between measures derived from dose-volume histograms and late rectal toxicities for radiotherapy of prostate cancer. The purpose of this study was to quantify correlations between measures describing the shape and location of the dose distribution and different outcomes. The dose to the rectal wall was projected on a two-dimensional map. In order to characterize the dose distribution, its centre of mass, longitudinal and lateral extent, and eccentricity were calculated at different dose levels. Furthermore, the dose-surface histogram (DSH) was determined. Correlations between these measures and seven clinically relevant rectal-toxicity endpoints were quantified by maximally selected standardized Wilcoxon rank statistics. The analysis was performed using data from the RT01 prostate radiotherapy trial. For some endpoints, the shape of the dose distribution is more strongly correlated with the outcome than simple DSHs. Rectal bleeding was most strongly correlated with the lateral extent of the dose distribution. For loose stools, the strongest correlations were found for longitudinal extent; proctitis was most strongly correlated with DSH. For the other endpoints no statistically significant correlations could be found. The strengths of the correlations between the shape of the dose distribution and outcome differed considerably between the different endpoints. Due to these significant correlations, it is desirable to use shape-based tools in order to assess the quality of a dose distribution.

Artificial neural networks for prediction of response to chemoradiation in HT29 xenografts

PURPOSE

To evaluate the feasibility of using neural networks for predicting treatment response by using longitudinal measurements of apparent diffusion coefficient (ADC) obtained from diffusion-weighted magnetic resonance imaging (DWMRI).

METHODS AND MATERIALS

Mice bearing HT29 xenografts were allocated to six treatment groups receiving different combinations of daily chemotherapy and/or radiation therapy for 2 weeks. T(2)-weighted and DWMR images were acquired before treatment, twice during fractionated chemoradiation (at days 4 and 11), and four times after treatment ended (at days 18, 25, 32, and 46). A tumor doubling growth delay (T(delay)) value was found for individual xenografts. ADC values and treatment groups (1-6) were used as input to a back propagation neural network (BPNN) to predict T(delay).

RESULTS

When treatment group and ADC values from days 0, 4, 11, 18, 25, 32, and 46 were used as inputs to the BPNN, a strong correlation between measured and predicted T(delay) values was found (R = 0.731, p < 0.01). When ADC values from days 0, 4, and 11, and the treatment group were used as inputs, the correlation between predicted and measured T(delay) was 0.693 (p < 0.01).

CONCLUSIONS

BPNN was successfully used to predict T(delay) from tumor ADC values obtained from HT29 xenografts undergoing fractionated chemoradiation therapy.

Predicting radiotherapy outcomes using statistical learning techniques

Radiotherapy outcomes are determined by complex interactions between treatment, anatomical and patient-related variables. A common obstacle to building maximally predictive outcome models for clinical practice is the failure to capture potential complexity of heterogeneous variable interactions and applicability beyond institutional data. We describe a statistical learning methodology that can automatically screen for nonlinear relations among prognostic variables and generalize to unseen data before. In this work, several types of linear and nonlinear kernels to generate interaction terms and approximate the treatment-response function are evaluated. Examples of institutional datasets of esophagitis, pneumonitis and xerostomia endpoints were used. Furthermore, an independent RTOG dataset was used for 'generalizabilty' validation. We formulated the discrimination between risk groups as a supervised learning problem. The distribution of patient groups was initially analyzed using principle components analysis (PCA) to uncover potential nonlinear behavior. The performance of the different methods was evaluated using bivariate correlations and actuarial analysis. Over-fitting was controlled via cross-validation resampling. Our results suggest that a modified support vector machine (SVM) kernel method provided superior performance on leave-one-out testing compared to logistic regression and neural networks in cases where the data exhibited nonlinear behavior on PCA. For instance, in prediction of esophagitis and pneumonitis endpoints, which exhibited nonlinear behavior on PCA, the method provided 21% and 60% improvements, respectively. Furthermore, evaluation on the independent pneumonitis RTOG dataset demonstrated good generalizabilty beyond institutional data in contrast with other models. This indicates that the prediction of treatment response can be improved by utilizing nonlinear kernel methods for discovering important nonlinear interactions among model variables. These models have the capacity to predict on unseen data.

Development of a set of nomograms to predict acute lower gastrointestinal toxicity for prostate cancer 3D-CRT

PURPOSE

To predict acute Radiation Therapy Oncology Group (RTOG)/European Organization for Research and Treatment of Cancer (EORTC) and Subjective Objective Signs Management and Analysis/Late Effect of Normal Tissue (SOMA/LENT) toxicities of the lower gastrointestinal (LGI) syndrome in patients with prostate cancer undergoing three-dimensional conformal radiotherapy using a tool (nomogram) that takes into account clinical and dosimetric variables that proved to be significant in the Italian Association for Radiation Oncology (AIRO) Group on Prostate Cancer (AIROPROS) 0102 trial.

METHODS AND MATERIALS

Acute rectal toxicity was scored in 1,132 patients by using both the RTOG/EORTC scoring system and a 10-item self-assessed questionnaire. Correlation between clinical variables/dose-volume histogram constraints and rectal toxicity was investigated by means of multivariate logistic analyses. Multivariate logistic analyses results were used to create nomograms predicting the symptoms of acute LGI syndrome.

RESULTS

Mean rectal dose was a strong predictor of Grade 2-3 RTOG/EORTC acute LGI toxicity (p = 0.0004; odds ratio (OR) = 1.035), together with hemorrhoids (p = 0.02; OR = 1.51), use of anticoagulants/antiaggregants (p = 0.02; OR = 0.63), and androgen deprivation (AD) (p = 0.04; OR = 0.65). Diabetes (p = 0.34; OR = 1.28) and pelvic node irradiation (p = 0.11; OR = 1.56) were significant variables to adjust toxicity prediction. Bleeding was related to hemorrhoids (p = 0.02; OR = 173), AD (p = 0.17; OR = 0.67), and mean rectal dose (p = 0.009; OR = 1.024). Stool frequency was related to seminal vesicle irradiation (p = 0.07; OR = 6.46), AD administered for more than 3 months (p = 0.002; OR = 0.32), and the percent volume of rectum receiving more than 60 Gy (V60Gy) V60 (p = 0.02; OR = 1.02). Severe fecal incontinence depended on seminal vesicle irradiation (p = 0.14; OR = 4.5) and V70 (p = 0.033; OR = 1.029).

CONCLUSIONS

To the best of our knowledge, this work presents the first set of nomograms available in the literature specific to symptoms of LGI syndrome and provides clinicians with a tailored probability of the specific outcome. Validation of the tool is in progress.

Application of artificial intelligence to the management of urological cancer

PURPOSE

Artificial intelligence techniques, such as artificial neural networks, Bayesian belief networks and neuro-fuzzy modeling systems, are complex mathematical models based on the human neuronal structure and thinking. Such tools are capable of generating data driven models of biological systems without making assumptions based on statistical distributions. A large amount of study has been reported of the use of artificial intelligence in urology. We reviewed the basic concepts behind artificial intelligence techniques and explored the applications of this new dynamic technology in various aspects of urological cancer management.

MATERIALS AND METHODS

A detailed and systematic review of the literature was performed using the MEDLINE and Inspec databases to discover reports using artificial intelligence in urological cancer.

RESULTS

The characteristics of machine learning and their implementation were described and reports of artificial intelligence use in urological cancer were reviewed. While most researchers in this field were found to focus on artificial neural networks to improve the diagnosis, staging and prognostic prediction of urological cancers, some groups are exploring other techniques, such as expert systems and neuro-fuzzy modeling systems.

CONCLUSIONS

Compared to traditional regression statistics artificial intelligence methods appear to be accurate and more explorative for analyzing large data cohorts. Furthermore, they allow individualized prediction of disease behavior. Each artificial intelligence method has characteristics that make it suitable for different tasks. The lack of transparency of artificial neural networks hinders global scientific community acceptance of this method but this can be overcome by neuro-fuzzy modeling systems.

Delayed disease progression after allogeneic cell vaccination in hormone-resistant prostate cancer and correlation with immunologic variables

PURPOSE

There are a significant number of patients with asymptomatic hormone-resistant prostate cancer who have increasing prostate-specific antigen (PSA) levels but little or no evaluable disease. The immunogenicity and minimal toxicity associated with cell-based vaccine therapy makes this approach attractive for these patients.

EXPERIMENTAL DESIGN

We have evaluated a vaccine comprising monthly intradermal injection of three irradiated allogeneic prostate cell lines (8 x 10(6) cells each) over 1 year. The first two doses were supplemented with bacille Calmette-Guérin as vaccine adjuvant. Twenty-eight hormone-resistant prostate cancer patients were enrolled. Patients were assessed clinically and PSA levels were measured monthly. Radiologic scans (X-ray, computed tomography, and bone scan) were taken at baseline and at intervals throughout the treatment period. Comprehensive monthly immunologic monitoring was undertaken including proliferation studies, activation markers, cytokine protein expression, and gene copy number. This longitudinal data was analyzed through predictive modeling using artificial neural network feed-forward/back-propagation algorithms with multilayer perceptron architecture.

RESULTS

Eleven of the 26 patients showed statistically significant, prolonged decreases in their PSA velocity (PSAV). None experienced any significant toxicity. Median time to disease progression was 58 weeks, compared with recent studies of other agents and historical control values of around 28 weeks. PSAV-responding patients showed a titratable T(H)1 cytokine release profile in response to restimulation with a vaccine lysate, while nonresponders showed a mixed T(H)1 and T(H)2 response. Furthermore, immunologic profile correlated with PSAV response by artificial neural network analysis. We found predictive power not only in expression of cytokines after maximal stimulation with phorbol 12-myristate 13-acetate, but also the method of analysis (qPCR measurement of IFN-gamma > qPCR measurement tumor necrosis factor-alpha > protein expression of IFN-gamma > protein expression of interleukin 2).

CONCLUSIONS

Whole cell allogeneic vaccination in hormone-resistant prostate cancer is nontoxic and improves the natural history of the disease. Longitudinal changes in immunologic function in vaccinated patients may be better interpreted through predictive modeling using tools such as the artificial neural network rather than periodic "snapshot" readouts.