Development of a robust radiomic biomarker of progression-free survival in advanced non-small cell lung cancer patients treated with first-line immunotherapy.

We aim to determine the feasibility of a novel radiomic biomarker that can integrate with other established clinical prognostic factors to predict progression-free survival (PFS) in patients with non-small cell lung cancer (NSCLC) undergoing first-line immunotherapy. Our study includes 107 patients with stage 4 NSCLC treated with pembrolizumab-based therapy (monotherapy: 30%, combination chemotherapy: 70%). The ITK-SNAP software was used for 3D tumor volume segmentation from pre-therapy CT scans. Radiomic features (n = 102) were extracted using the CaPTk software. Impact of heterogeneity introduced by image physical dimensions (voxel spacing parameters) and acquisition parameters (contrast enhancement and CT reconstruction kernel) was mitigated by resampling the images to the minimum voxel spacing parameters and harmonization by a nested ComBat technique. This technique was initialized with radiomic features, clinical factors of age, sex, race, PD-L1 expression, ECOG status, body mass index (BMI), smoking status, recurrence event and months of progression-free survival, and image acquisition parameters as batch variables. Two phenotypes were identified using unsupervised hierarchical clustering of harmonized features. Prognostic factors, including PDL1 expression, ECOG status, BMI and smoking status, were combined with radiomic phenotypes in Cox regression models of PFS and Kaplan Meier (KM) curve-fitting. Cox model based on clinical factors had a c-statistic of 0.57, which increased to 0.63 upon addition of phenotypes derived from harmonized features. There were statistically significant differences in survival outcomes stratified by clinical covariates, as measured by the log-rank test (p = 0.034), which improved upon addition of phenotypes (p = 0.00022). We found that mitigation of heterogeneity by image resampling and nested ComBat harmonization improves prognostic value of phenotypes, resulting in better prediction of PFS when added to other prognostic variables.

Radiomic Phenotypes for Improving Early Prediction of Survival in Stage III Non-Small Cell Lung Cancer Adenocarcinoma after Chemoradiation.

We evaluate radiomic phenotypes derived from CT scans as early predictors of overall survival (OS) after chemoradiation in stage III primary lung adenocarcinoma. We retrospectively analyzed 110 thoracic CT scans acquired between April 2012-October 2018. Patients received a median radiation dose of 66.6 Gy at 1.8 Gy/fraction delivered with proton (55.5%) and photon (44.5%) beam treatment, as well as concurrent chemotherapy (89%) with carboplatin-based (55.5%) and cisplatin-based (36.4%) doublets. A total of 56 death events were recorded. Using manual tumor segmentations, 107 radiomic features were extracted. Feature harmonization using ComBat was performed to mitigate image heterogeneity due to the presence or lack of intravenous contrast material and variability in CT scanner vendors. A binary radiomic phenotype to predict OS was derived through the unsupervised hierarchical clustering of the first principal components explaining 85% of the variance of the radiomic features. C-scores and likelihood ratio tests (LRT) were used to compare the performance of a baseline Cox model based on ECOG status and age, with a model integrating the radiomic phenotype with such clinical predictors. The model integrating the radiomic phenotype (C-score = 0.69, 95% CI = (0.62, 0.77)) significantly improved (p<0.005) upon the baseline model (C-score = 0.65, CI = (0.57, 0.73)). Our results suggest that harmonized radiomic phenotypes can significantly improve OS prediction in stage III NSCLC after chemoradiation.

Impact of Interobserver Variability in Manual Segmentation of Non-Small Cell Lung Cancer (NSCLC) Applying Low-Rank Radiomic Representation on Computed Tomography.

This study tackles interobserver variability with respect to specialty training in manual segmentation of non-small cell lung cancer (NSCLC). Four readers included for segmentation are: a data scientist (BY), a medical student (LS), a radiology trainee (MH), and a specialty-trained radiologist (SK) for a total of 293 patients from two publicly available databases. Sørensen-Dice (SD) coefficients and low rank Pearson correlation coefficients (CC) of 429 radiomics were calculated to assess interobserver variability. Cox proportional hazard (CPH) models and Kaplan-Meier (KM) curves of overall survival (OS) prediction for each dataset were also generated. SD and CC for segmentations demonstrated high similarities, yielding, SD: 0.79 and CC: 0.92 (BY-SK), SD: 0.81 and CC: 0.83 (LS-SK), and SD: 0.84 and CC: 0.91 (MH-SK) in average for both databases, respectively. OS through the maximal CPH model for the two datasets yielded c-statistics of 0.7 (95% CI) and 0.69 (95% CI), while adding radiomic and clinical variables (sex, stage/morphological status, and histology) together. KM curves also showed significant discrimination between high- and low-risk patients (p-value < 0.005). This supports that readers' level of training and clinical experience may not significantly influence the ability to extract accurate radiomic features for NSCLC on CT. This potentially allows flexibility in the training required to produce robust prognostic imaging biomarkers for potential clinical translation.

Machine learning highlights the deficiency of conventional dosimetric constraints for prevention of high-grade radiation esophagitis in non-small cell lung cancer treated with chemoradiation.

BACKGROUND AND PURPOSE: Radiation esophagitis is a clinically important toxicity seen with treatment for locally-advanced non-small cell lung cancer. There is considerable disagreement among prior studies in identifying predictors of radiation esophagitis. We apply machine learning algorithms to identify factors contributing to the development of radiation esophagitis to uncover previously unidentified criteria and more robust dosimetric factors. MATERIALS AND METHODS: We used machine learning approaches to identify predictors of grade ≥ 3 radiation esophagitis in a cohort of 202 consecutive locally-advanced non-small cell lung cancer patients treated with definitive chemoradiation from 2008 to 2016. We evaluated 35 clinical features per patient grouped into risk factors, comorbidities, imaging, stage, histology, radiotherapy, chemotherapy and dosimetry. Univariate and multivariate analyses were performed using a panel of 11 machine learning algorithms combined with predictive power assessments. RESULTS: All patients were treated to a median dose of 66.6 Gy at 1.8 Gy per fraction using photon (89.6%) and proton (10.4%) beam therapy, most often with concurrent chemotherapy (86.6%). 11.4% of patients developed grade ≥ 3 radiation esophagitis. On univariate analysis, no individual feature was found to predict radiation esophagitis (AUC range 0.45-0.55, p ≥ 0.07). In multivariate analysis, all machine learning algorithms exhibited poor predictive performance (AUC range 0.46-0.56, p ≥ 0.07). CONCLUSIONS: Contemporary machine learning algorithms applied to our modern, relatively large institutional cohort could not identify any reliable predictors of grade ≥ 3 radiation esophagitis. Additional patients are needed, and novel patient-specific and treatment characteristics should be investigated to develop clinically meaningful methods to mitigate this survival altering toxicity.

Expert-augmented machine learning.

Machine learning is proving invaluable across disciplines. However, its success is often limited by the quality and quantity of available data, while its adoption is limited by the level of trust afforded by given models. Human vs. machine performance is commonly compared empirically to decide whether a certain task should be performed by a computer or an expert. In reality, the optimal learning strategy may involve combining the complementary strengths of humans and machines. Here, we present expert-augmented machine learning (EAML), an automated method that guides the extraction of expert knowledge and its integration into machine-learned models. We used a large dataset of intensive-care patient data to derive 126 decision rules that predict hospital mortality. Using an online platform, we asked 15 clinicians to assess the relative risk of the subpopulation defined by each rule compared to the total sample. We compared the clinician-assessed risk to the empirical risk and found that, while clinicians agreed with the data in most cases, there were notable exceptions where they overestimated or underestimated the true risk. Studying the rules with greatest disagreement, we identified problems with the training data, including one miscoded variable and one hidden confounder. Filtering the rules based on the extent of disagreement between clinician-assessed risk and empirical risk, we improved performance on out-of-sample data and were able to train with less data. EAML provides a platform for automated creation of problem-specific priors, which help build robust and dependable machine-learning models in critical applications.

Building more accurate decision trees with the additive tree.

The expansion of machine learning to high-stakes application domains such as medicine, finance, and criminal justice, where making informed decisions requires clear understanding of the model, has increased the interest in interpretable machine learning. The widely used Classification and Regression Trees (CART) have played a major role in health sciences, due to their simple and intuitive explanation of predictions. Ensemble methods like gradient boosting can improve the accuracy of decision trees, but at the expense of the interpretability of the generated model. Additive models, such as those produced by gradient boosting, and full interaction models, such as CART, have been investigated largely in isolation. We show that these models exist along a spectrum, revealing previously unseen connections between these approaches. This paper introduces a rigorous formalization for the additive tree, an empirically validated learning technique for creating a single decision tree, and shows that this method can produce models equivalent to CART or gradient boosted stumps at the extremes by varying a single parameter. Although the additive tree is designed primarily to provide both the model interpretability and predictive performance needed for high-stakes applications like medicine, it also can produce decision trees represented by hybrid models between CART and boosted stumps that can outperform either of these approaches.

Predicting radiation pneumonitis in locally advanced stage II-III non-small cell lung cancer using machine learning.

BACKGROUND AND PURPOSE: Radiation pneumonitis (RP) is a radiotherapy dose-limiting toxicity for locally advanced non-small cell lung cancer (LA-NSCLC). Prior studies have proposed relevant dosimetric constraints to limit this toxicity. Using machine learning algorithms, we performed analyses of contributing factors in the development of RP to uncover previously unidentified criteria and elucidate the relative importance of individual factors. MATERIALS AND METHODS: We evaluated 32 clinical features per patient in a cohort of 203 stage II-III LA-NSCLC patients treated with definitive chemoradiation to a median dose of 66.6 Gy in 1.8 Gy daily fractions at our institution from 2008 to 2016. Of this cohort, 17.7% of patients developed grade ≥2 RP. Univariate analysis was performed using trained decision stumps to individually analyze statistically significant predictors of RP and perform feature selection. Applying Random Forest, we performed multivariate analysis to assess the combined performance of important predictors of RP. RESULTS: On univariate analysis, lung V20, lung mean, lung V10 and lung V5 were found to be significant RP predictors with the greatest balance of specificity and sensitivity. On multivariate analysis, Random Forest (AUC = 0.66, p = 0.0005) identified esophagus max (20.5%), lung V20 (16.4%), lung mean (15.7%) and pack-year (14.9%) as the most common primary differentiators of RP. CONCLUSIONS: We highlight Random Forest as an accurate machine learning method to identify known and new predictors of symptomatic RP. Furthermore, this analysis confirms the importance of lung V20, lung mean and pack-year as predictors of RP while also introducing esophagus max as an important RP predictor.

MediBoost: a Patient Stratification Tool for Interpretable Decision Making in the Era of Precision Medicine.

Machine learning algorithms that are both interpretable and accurate are essential in applications such as medicine where errors can have a dire consequence. Unfortunately, there is currently a tradeoff between accuracy and interpretability among state-of-the-art methods. Decision trees are interpretable and are therefore used extensively throughout medicine for stratifying patients. Current decision tree algorithms, however, are consistently outperformed in accuracy by other, less-interpretable machine learning models, such as ensemble methods. We present MediBoost, a novel framework for constructing decision trees that retain interpretability while having accuracy similar to ensemble methods, and compare MediBoost's performance to that of conventional decision trees and ensemble methods on 13 medical classification problems. MediBoost significantly outperformed current decision tree algorithms in 11 out of 13 problems, giving accuracy comparable to ensemble methods. The resulting trees are of the same type as decision trees used throughout clinical practice but have the advantage of improved accuracy. Our algorithm thus gives the best of both worlds: it grows a single, highly interpretable tree that has the high accuracy of ensemble methods.