Gilmer Valdes

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.

A mathematical framework for virtual IMRT QA using machine learning.

PURPOSE It is common practice to perform patient-specific pretreatment verifications to the clinical delivery of IMRT. This process can be time-consuming and not altogether instructive due to the myriad sources that may produce a failing result. The purpose of this study was to develop an algorithm capable of predicting IMRT QA passing rates a priori. METHODS From all treatment, 498 IMRT plans sites were planned in eclipse version 11 and delivered using a dynamic sliding window technique on Clinac iX or TrueBeam Linacs. 3%/3 mm local dose/distance-to-agreement (DTA) was recorded using a commercial 2D diode array. Each plan was characterized by 78 metrics that describe different aspects of their complexity that could lead to disagreements between the calculated and measured dose. A Poisson regression with Lasso regularization was trained to learn the relation between the plan characteristics and each passing rate. RESULTS Passing rates 3%/3 mm local dose/DTA can be predicted with an error smaller than 3% for all plans analyzed. The most important metrics to describe the passing rates were determined to be the MU factor (MU per Gy), small aperture score, irregularity factor, and fraction of the plan delivered at the corners of a 40 × 40 cm field. The higher the value of these metrics, the worse the passing rates. CONCLUSIONS The Virtual QA process predicts IMRT passing rates with a high likelihood, allows the detection of failures due to setup errors, and it is sensitive enough to detect small differences between matched Linacs.

IMRT QA using machine learning: A multi‐institutional validation

Abstract Purpose To validate a machine learning approach to Virtual intensity‐modulated radiation therapy (IMRT) quality assurance (QA) for accurately predicting gamma passing rates using different measurement approaches at different institutions. Methods A Virtual IMRT QA framework was previously developed using a machine learning algorithm based on 498 IMRT plans, in which QA measurements were performed using diode‐array detectors and a 3%local/3 mm with 10% threshold at Institution 1. An independent set of 139 IMRT measurements from a different institution, Institution 2, with QA data based on portal dosimetry using the same gamma index, was used to test the mathematical framework. Only pixels with ≥10% of the maximum calibrated units (CU) or dose were included in the comparison. Plans were characterized by 90 different complexity metrics. A weighted poison regression with Lasso regularization was trained to predict passing rates using the complexity metrics as input. Results The methodology predicted passing rates within 3% accuracy for all composite plans measured using diode‐array detectors at Institution 1, and within 3.5% for 120 of 139 plans using portal dosimetry measurements performed on a per‐beam basis at Institution 2. The remaining measurements (19) had large areas of low CU, where portal dosimetry has a larger disagreement with the calculated dose and as such, the failure was expected. These beams need further modeling in the treatment planning system to correct the under‐response in low‐dose regions. Important features selected by Lasso to predict gamma passing rates were as follows: complete irradiated area outline (CIAO), jaw position, fraction of MLC leafs with gaps smaller than 20 or 5 mm, fraction of area receiving less than 50% of the total CU, fraction of the area receiving dose from penumbra, weighted average irregularity factor, and duty cycle. Conclusions We have demonstrated that Virtual IMRT QA can predict passing rates using different measurement techniques and across multiple institutions. Prediction of QA passing rates can have profound implications on the current IMRT process.

Radiosensitization of gliomas by intracellular generation of 5-fluorouracil potentiates prodrug activator gene therapy with a retroviral replicating vector

A tumor-selective non-lytic retroviral replicating vector (RRV), Toca 511, and an extended-release formulation of 5-fluorocytosine (5-FC), Toca FC, are currently being evaluated in clinical trials in patients with recurrent high-grade glioma (NCT01156584, NCT01470794 and NCT01985256). Tumor-selective propagation of this RRV enables highly efficient transduction of glioma cells with cytosine deaminase (CD), which serves as a prodrug activator for conversion of the anti-fungal prodrug 5-FC to the anti-cancer drug 5-fluorouracil (5-FU) directly within the infected cells. We investigated whether, in addition to its direct cytotoxic effects, 5-FU generated intracellularly by RRV-mediated CD/5-FC prodrug activator gene therapy could also act as a radiosensitizing agent. Efficient transduction by RRV and expression of CD were confirmed in the highly aggressive, radioresistant human glioblastoma cell line U87EGFRvIII and its parental cell line U87MG (U87). RRV-transduced cells showed significant radiosensitization even after transient exposure to 5-FC. This was confirmed both in vitro by a clonogenic colony survival assay and in vivo by bioluminescence imaging analysis. These results provide a convincing rationale for development of tumor-targeted radiosensitization strategies utilizing the tumor-selective replicative capability of RRV, and incorporation of radiation therapy into future clinical trials evaluating Toca 511 and Toca FC in brain tumor patients.

Tumor control probability and the utility of 4D vs 3D dose calculations for stereotactic body radiotherapy for lung cancer

Four-dimensional (4D) dose calculations for lung cancer radiotherapy have been technically feasible for a number of years but have not become standard clinical practice. The purpose of this study was to determine if clinically significant differences in tumor control probability (TCP) exist between 3D and 4D dose calculations so as to inform the decision whether 4D dose calculations should be used routinely for treatment planning. Radiotherapy plans for Stage I-II lung cancer were created for 8 patients. Clinically acceptable treatment plans were created with dose calculated on the end-exhale 4D computed tomography (CT) phase using a Monte Carlo algorithm. Dose was then projected onto the remaining 9 phases of 4D-CT using the Monte Carlo algorithm and accumulated onto the end-exhale phase using commercially available deformable registration software. The resulting dose-volume histograms (DVH) of the gross tumor volume (GTV), planning tumor volume (PTV), and PTVsetup were compared according to target coverage and dose. The PTVsetup was defined as a volume including the GTV and a margin for setup uncertainties but not for respiratory motion. TCPs resulting from these DVHs were estimated using a wide range of alphas, betas, and tumor cell densities. Differences of up to 5Gy were observed between 3D and 4D calculations for a PTV with highly irregular shape. When the TCP was calculated using the resulting DVHs for fractionation schedules typically used in stereotactic body radiation therapy (SBRT), the TCP differed at most by 5% between 4D and 3D cases, and in most cases, it was by less than 1%. We conclude that 4D dose calculations are not necessary for most cases treated with SBRT, but they might be valuable for irregularly shaped target volumes. If 4D calculations are used, 4D DVHs should be evaluated on volumes that include margin for setup uncertainty but not respiratory motion.

The High-Affinity Maltose Switch MBP317-347 has Low Affinity for Glucose: Implications for Targeting Tumors with Metabolically Directed Enzyme Prodrug Therapy

Development of agents with high affinity and specificity for tumor‐specific markers is an important goal of molecular‐targeted therapy. Here, we propose a shift in paradigm using a strategy that relies on low affinity for fundamental metabolites found in different concentrations in cancerous and non‐cancerous tissues: glucose and lactate. A molecular switch, MBP317‐347, originally designed to be a high‐affinity switch for maltose and maltose‐like polysaccharides, was demonstrated to be a low‐affinity switch for glucose, that is, able to be activated by high concentrations (tens of millimolar) of glucose. We propose that such a low‐affinity glucose switch could be used as a proof of concept for a new prodrug therapy strategy denominated metabolically directed enzyme prodrug therapy (MDEPT) where glucose or, preferably, lactate serves as the activator. Accordingly, considering the typical differential concentrations of lactate found in tumors and in healthy tissues, a low‐affinity lactate‐binding switch analogous to the low‐affinity glucose‐binding switch MBP317‐347 would be an order of magnitude more active in tumors than in normal tissues and therefore can work as a differential activator of anticancer drugs in tumors.

Re-evaluation of cellular radiosensitization by 5-fluorouracil: High-dose, pulsed administration is effective and preferable to conventional low-dose, chronic administration

Abstract Purpose: It is widely believed that the anticancer drug 5- fluorouracil (5-FU) must be administered chronically and in low doses to maximize radiosensitization during chemoradiotherapy. The rationale is based upon cell experiments that assumed identical mechanisms of 5-FU action between low-dose chronic (LDC) and high-dose pulsed (HDP) exposures. Here we challenge the paradigm and demonstrate the effectiveness of HDP 5-FU as a radiosensitizer and the wide range of dose/time schedules that can be used to synergize with radiation as compared to the relatively restrictive protocols prescribed for current LDC administrations. Materials and methods: Clonogenic survival of human glioblastoma and colon cancer cell lines, U87MG-VIII and HCT-116, respectively, was used to assess temporal and dose effects of 5-FU on radiosensitivity and in split-dose experiments to characterize changes in sublethal damage repair. Results: We show that HDP 5-FU administration does indeed radiosensitize both the highly radioresistant U87MG-VIII and HCT-116. Additionally, we show that this radiosensitization lasts for at least 24 h if cells are pre-irradiated with 2 Gy immediately after HDP 5-FU exposure as a result of a decrease in sublethal damage repair capacity for subsequent irradiations, suggesting the ideal combination of 5-FU bolus injection with fractionation radiotherapy schemes. Conclusions: 5-FU bolus administration protocols combined with radiation would not only help improve treatment outcomes and reduce development of 5-FU resistance, but it would greatly benefit patients by shortening clinical stays and lowering overall therapeutic costs.

Machine Learning in Radiation Oncology: Opportunities, Requirements, and Needs

Machine learning (ML) has the potential to revolutionize the field of radiation oncology, but there is much work to be done. In this article, we approach the radiotherapy process from a workflow perspective, identifying specific areas where a data-centric approach using ML could improve the quality and efficiency of patient care. We highlight areas where ML has already been used, and identify areas where we should invest additional resources. We believe that this article can serve as a guide for both clinicians and researchers to start discussing issues that must be addressed in a timely manner.

An unsupervised convolutional neural network-based algorithm for deformable image registration

The purpose of the work is to develop a deep unsupervised learning strategy for cone-beam CT (CBCT) to CT deformable image registration (DIR). This technique uses a deep convolutional inverse graphics network (DCIGN) based DIR algorithm implemented on 2 Nvidia 1080 Ti graphics processing units. The model is comprised of an encoding and decoding stage. The fully-convolutional encoding stage learns hierarchical features and simultaneously forms an information bottleneck, while the decoding stage restores the original dimensionality of the input image. Activations from the encoding stage are used as the input channels to a sparse DIR algorithm. DCIGN was trained using a distributive learning-based convolutional neural network architecture and used 285 head and neck patients to train, validate, and test the algorithm. The accuracy of the DCIGN algorithm was evaluated on 100 synthetic cases and 12 hold out test patient cases. The results indicate that DCIGN performed better than rigid registration, intensity corrected Demons, and landmark-guided deformable image registration for all evaluation metrics. DCIGN required ~14 h to train, and ~3.5 s to make a prediction on a 512  ×  512  ×  120 voxel image. In conclusion, DCIGN is able to maintain high accuracy in the presence of CBCT noise contamination, while simultaneously preserving high computational efficiency.

Preoperative and postoperative prediction of long-term meningioma outcomes

Background Meningiomas are stratified according to tumor grade and extent of resection, often in isolation of other clinical variables. Here, we use machine learning (ML) to integrate demographic, clinical, radiographic and pathologic data to develop predictive models for meningioma outcomes. Methods and findings We developed a comprehensive database containing information from 235 patients who underwent surgery for 257 meningiomas at a single institution from 1990 to 2015. The median follow-up was 4.3 years, and resection specimens were re-evaluated according to current diagnostic criteria, revealing 128 WHO grade I, 104 grade II and 25 grade III meningiomas. A series of ML algorithms were trained and tuned by nested resampling to create models based on preoperative features, conventional postoperative features, or both. We compared different algorithms’ accuracy as well as the unique insights they offered into the data. Machine learning models restricted to preoperative information, such as patient demographics and radiographic features, had similar accuracy for predicting local failure (AUC = 0.74) or overall survival (AUC = 0.68) as models based on meningioma grade and extent of resection (AUC = 0.73 and AUC = 0.72, respectively). Integrated models incorporating all available demographic, clinical, radiographic and pathologic data provided the most accurate estimates (AUC = 0.78 and AUC = 0.74, respectively). From these models, we developed decision trees and nomograms to estimate the risks of local failure or overall survival for meningioma patients. Conclusions Clinical information has been historically underutilized in the prediction of meningioma outcomes. Predictive models trained on preoperative clinical data perform comparably to conventional models trained on meningioma grade and extent of resection. Combination of all available information can help stratify meningioma patients more accurately.