Statistical biopsy: An emerging screening approach for early detection of cancers.

Despite large investment cancer continues to be a major source of mortality and morbidity throughout the world. Traditional methods of detection and diagnosis such as biopsy and imaging, tend to be expensive and have risks of complications. As data becomes more abundant and machine learning continues advancing, it is natural to ask how they can help solve some of these problems. In this paper we show that using a person's personal health data it is possible to predict their risk for a wide variety of cancers. We dub this process a "statistical biopsy." Specifically, we train two neural networks, one predicting risk for 16 different cancer types in females and the other predicting risk for 15 different cancer types in males. The networks were trained as binary classifiers identifying individuals that were diagnosed with the different cancer types within 5 years of joining the PLOC trial. However, rather than use the binary output of the classifiers we show that the continuous output can instead be used as a cancer risk allowing a holistic look at an individual's cancer risks. We tested our multi-cancer model on the UK Biobank dataset showing that for most cancers the predictions generalized well and that looking at multiple cancer risks at once from personal health data is a possibility. While the statistical biopsy will not be able to replace traditional biopsies for diagnosing cancers, we hope there can be a shift of paradigm in how statistical models are used in cancer detection moving to something more powerful and more personalized than general population screening guidelines.

A general-purpose Monte Carlo particle transport code based on inverse transform sampling for radiotherapy dose calculation.

The Monte Carlo (MC) method is widely used to solve various problems in radiotherapy. There has been an impetus to accelerate MC simulation on GPUs whereas thread divergence remains a major issue for MC codes based on acceptance-rejection sampling. Inverse transform sampling has the potential to eliminate thread divergence but it is only implemented for photon transport. Here, we report a MC package Particle Transport in Media (PTM) to demonstrate the implementation of coupled photon-electron transport simulation using inverse transform sampling. Rayleigh scattering, Compton scattering, photo-electric effect and pair production are considered in an analogous manner for photon transport. Electron transport is simulated in a class II condensed history scheme, i.e., catastrophic inelastic scattering and Bremsstrahlung events are simulated explicitly while subthreshold interactions are subject to grouping. A random-hinge electron step correction algorithm and a modified PRESTA boundary crossing algorithm are employed to improve simulation accuracy. Benchmark studies against both EGSnrc simulations and experimental measurements are performed for various beams, phantoms and geometries. Gamma indices of the dose distributions are better than 99.6% for all the tested scenarios under the 2%/2 mm criteria. These results demonstrate the successful implementation of inverse transform sampling in coupled photon-electron transport simulation.

Monte Carlo simulation of coherently scattered photons based on the inverse-sampling technique.

The acceptance-rejection technique has been widely used in several Monte Carlo simulation packages for Rayleigh scattering of photons. However, the models implemented in these packages might fail to reproduce the corresponding experimental and theoretical results. The discrepancy is attributed to the fact that all current simulations implement an elastic scattering model for the angular distribution of photons without considering anomalous scattering effects. In this study, a novel Rayleigh scattering model using anomalous scattering factors based on the inverse-sampling technique is presented. Its performance was evaluated against other simulation algorithms in terms of simulation accuracy and computational efficiency. The computational efficiency was tested with a general-purpose Monte Carlo package named Particle Transport in Media (PTM). The evaluation showed that a Monte Carlo model using both atomic form factors and anomalous scattering factors for the angular distribution of photons (instead of the atomic form factors alone) produced Rayleigh scattering results in closer agreement with experimental data. The comparison and evaluation confirmed that the inverse-sampling technique using atomic form factors and anomalous scattering factors exhibited improved computational efficiency and performed the best in reproducing experimental measurements and related scattering matrix calculations. Furthermore, using this model to sample coherent scattering can provide scientific insight for complex systems.

Robust Machine Learning for Colorectal Cancer Risk Prediction and Stratification.

While colorectal cancer (CRC) is third in prevalence and mortality among cancers in the United States, there is no effective method to screen the general public for CRC risk. In this study, to identify an effective mass screening method for CRC risk, we evaluated seven supervised machine learning algorithms: linear discriminant analysis, support vector machine, naive Bayes, decision tree, random forest, logistic regression, and artificial neural network. Models were trained and cross-tested with the National Health Interview Survey (NHIS) and the Prostate, Lung, Colorectal, Ovarian Cancer Screening (PLCO) datasets. Six imputation methods were used to handle missing data: mean, Gaussian, Lorentzian, one-hot encoding, Gaussian expectation-maximization, and listwise deletion. Among all of the model configurations and imputation method combinations, the artificial neural network with expectation-maximization imputation emerged as the best, having a concordance of 0.70 ± 0.02, sensitivity of 0.63 ± 0.06, and specificity of 0.82 ± 0.04. In stratifying CRC risk in the NHIS and PLCO datasets, only 2% of negative cases were misclassified as high risk and 6% of positive cases were misclassified as low risk. In modeling the CRC-free probability with Kaplan-Meier estimators, low-, medium-, and high CRC-risk groups have statistically-significant separation. Our results indicated that the trained artificial neural network can be used as an effective screening tool for early intervention and prevention of CRC in large populations.

Population-Based Screening for Endometrial Cancer: Human vs. Machine Intelligence.

Incidence and mortality rates of endometrial cancer are increasing, leading to increased interest in endometrial cancer risk prediction and stratification to help in screening and prevention. Previous risk models have had moderate success with the area under the curve (AUC) ranging from 0.68 to 0.77. Here we demonstrate a population-based machine learning model for endometrial cancer screening that achieves a testing AUC of 0.96. We train seven machine learning algorithms based solely on personal health data, without any genomic, imaging, biomarkers, or invasive procedures. The data come from the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO). We further compare our machine learning model with 15 gynecologic oncologists and primary care physicians in the stratification of endometrial cancer risk for 100 women. We find a random forest model that achieves a testing AUC of 0.96 and a neural network model that achieves a testing AUC of 0.91. We test both models in risk stratification against 15 practicing physicians. Our random forest model is 2.5 times better at identifying above-average risk women with a 2-fold reduction in the false positive rate. Our neural network model is 2 times better at identifying above-average risk women with a 3-fold reduction in the false positive rate. Our machine learning models provide a non-invasive and cost-effective way to identify high-risk sub-populations who may benefit from early screening of endometrial cancer, prior to disease onset. Through statistical biopsy of personal health data, we have identified a new and effective approach for early cancer detection and prevention for individual patients.

Predicting breast cancer risk using personal health data and machine learning models.

Among women, breast cancer is a leading cause of death. Breast cancer risk predictions can inform screening and preventative actions. Previous works found that adding inputs to the widely-used Gail model improved its ability to predict breast cancer risk. However, these models used simple statistical architectures and the additional inputs were derived from costly and / or invasive procedures. By contrast, we developed machine learning models that used highly accessible personal health data to predict five-year breast cancer risk. We created machine learning models using only the Gail model inputs and models using both Gail model inputs and additional personal health data relevant to breast cancer risk. For both sets of inputs, six machine learning models were trained and evaluated on the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial data set. The area under the receiver operating characteristic curve metric quantified each model's performance. Since this data set has a small percentage of positive breast cancer cases, we also reported sensitivity, specificity, and precision. We used Delong tests (p < 0.05) to compare the testing data set performance of each machine learning model to that of the Breast Cancer Risk Prediction Tool (BCRAT), an implementation of the Gail model. None of the machine learning models with only BCRAT inputs were significantly stronger than the BCRAT. However, the logistic regression, linear discriminant analysis, and neural network models with the broader set of inputs were all significantly stronger than the BCRAT. These results suggest that relative to the BCRAT, additional easy-to-obtain personal health inputs can improve five-year breast cancer risk prediction. Our models could be used as non-invasive and cost-effective risk stratification tools to increase early breast cancer detection and prevention, motivating both immediate actions like screening and long-term preventative measures such as hormone replacement therapy and chemoprevention.

Scoring colorectal cancer risk with an artificial neural network based on self-reportable personal health data.

Colorectal cancer (CRC) is third in prevalence and mortality among all cancers in the US. Currently, the United States Preventative Services Task Force (USPSTF) recommends anyone ages 50-75 and/or with a family history to be screened for CRC. To improve screening specificity and sensitivity, we have built an artificial neural network (ANN) trained on 12 to 14 categories of personal health data from the National Health Interview Survey (NHIS). Years 1997-2016 of the NHIS contain 583,770 respondents who had never received a diagnosis of any cancer and 1409 who had received a diagnosis of CRC within 4 years of taking the survey. The trained ANN has sensitivity of 0.57 ± 0.03, specificity of 0.89 ± 0.02, positive predictive value of 0.0075 ± 0.0003, negative predictive value of 0.999 ± 0.001, and concordance of 0.80 ± 0.05 per the guidelines of Transparent Reporting of Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) level 2a, comparable to current risk-scoring methods. To demonstrate clinical applicability, both USPSTF guidelines and the trained ANN are used to stratify respondents to the 2017 NHIS into low-, medium- and high-risk categories (TRIPOD levels 4 and 2b, respectively). The number of CRC respondents misclassified as low risk is decreased from 35% by screening guidelines to 5% by ANN (in 60 cases). The number of non-CRC respondents misclassified as high risk is decreased from 53% by screening guidelines to 6% by ANN (in 25,457 cases). Our results demonstrate a robustly-tested method of stratifying CRC risk that is non-invasive, cost-effective, and easy to implement publicly.

Stratifying Ovarian Cancer Risk Using Personal Health Data.

Purpose: Screening the general population for ovarian cancer is not recommended by every major medical or public health organization because the harms from screening outweigh the benefit it provides. To improve ovarian cancer detection and survival many are looking at high-risk populations who would benefit from screening. Methods: We train a neural network on readily available personal health data to predict and stratify ovarian cancer risk. We use two different datasets to train our network: The National Health Interview Survey and Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial. Results: Our model has an area under the receiver operating characteristic curve of 0.71. We further demonstrate how the model could be used to stratify patients into different risk categories. A simple 3-tier scheme classifies 23.8% of those with cancer and 1.0% of those without as high-risk similar to genetic testing, and 1.1% of those with cancer and 24.4% of those without as low risk. Conclusion: The developed neural network offers a cost-effective and non-invasive way to identify those who could benefit from targeted screening.