Espadon, an R package for automation, exploitation and processing of DICOM files in medical physics and clinical research.

INTRODUCTION: One of the main issues in the field of clinical research is to enhance clinical databases with information from imaging (CT, MR, PET-scan), contouring (RTstruct), or produced by TPS such as dose distribution (RTdose) or treatment plans (RTplan). To perform these analyses automatically, we propose the new open-source package "espadon", developed in R environment. This package also opens up numerous perspectives for TPS-independant calculation, automation and processing of DICOM data. RESULTS: The espadon package converts DICOM objects into espadon objects. Several tools have been developed to manipulate these objects and extract the desired information. In addition to decode DICOM files and pseudonomize them, the great advantage of espadon is that it presents the links between patient data (images, structures, treatment plans) in a didactic way, respecting the dates of the examinations. It can visualize volumes or structures in 2D or 3D, resample volumes, segment them, and change geometric frames of reference. It integrates dose-volume histogram functions on a selection, with Monte Carlo calculations of random shifts of contours. It offers the automatic calculation of several usual radiotherapy indices, as well as the calculation of Gamma and Chi indices. CONCLUSIONS: Espadon is a toolkit designed to be easily used by radiotherapists, medical physicists or students. Espadon's functions are implemented in an R script, and allow the automatic extraction or calculation of data from DICOM files, which can be used for statistical modelling or machine-learning in the R environment. This package is available on the Comprehensive R Archive Network (CRAN) repository.

A Comprehensive Analysis of the Relationship Between Dose Rate and Biological Effects in Preclinical and Clinical Studies, From Brachytherapy to Flattening Filter Free Radiation Therapy and FLASH Irradiation.

PURPOSE: For many years, the effect of dose rate (DR) was considered negligible in external beam radiation therapy (EBRT) until very-high DR (>10 Gy/min) became possible and ultrahigh DR (>40 Gy/s) showed dramatic protection of normal tissues in preclinical experiments. We propose a critical review of preclinical and clinical studies to investigate the biological and clinical effects of DR variation in the range covering brachytherapy to flattening filter free EBRT and FLASH. METHODS AND MATERIALS: Preclinical and clinical studies investigating biological and clinical DR effects were reviewed extensively. We also conducted an in silico study to assess the effect of pulse DR (DRp), taking into account the mean time between 2 tracks during the pulse. RESULTS: Preclinical studies have shown that an increase in DR in the range of 0.01 to 20 Gy/min (not including ultralow or ultrahigh DR) resulted in decreased survival of both normal and tumor cells. This effect was attributed primarily to increasingly unrepaired "sublethal" DNA damage with increasing the DR. However, the models and irradiation conditions have often been very different from one radiobiological study to another. Moreover, the physical parameters on the spatial and temporal microstructure of the beam were not considered systematically. In particular, the DRp was rarely mentioned. The in silico studies showed that for the same average DR, increasing DRp induced an increase of mean track rates. These results could explain the presence of more complex damage when the DRp was increased within the range of DR considered, in relation to the time-dependent probability of accumulating unrepaired, "sublethal" DNA lesions in close proximity. CONCLUSIONS: Knowledge of the beam microstructure is critical to understanding the biological impact and the clinical outcomes of radiation at the DR commonly used in radiation therapy.

Development of clinically based prediction models using machine learning and Bayesian statistics / Desarrollo de modelos de predicción basados clínicamente utilizando aprendizaje automático y estadísticas bayesianas

Abstract In this work, the framework for developing generic clinically based models is emphasized and illustrated with Bayesian statistics neurologic grade prediction models in order to exemplify the type of models that can be developed from a mathematical point of view. The models are based on clinical records of patients who underwent radiotherapy treatment due to glioblastoma which is an aggressive brain cancer. A first model requires as a parameter the neurologic grade of the patient before the treatment then predicts the grade after the treatment. A second, enhanced, model was developed with the aim of making the prediction more realistic and it uses the neurologic grade before the treatment as well, but it additionally depends on the Clinical Target Volume (CTV). Furthermore, with the aid of Bayesian statistic we were able to estimate the uncertainty of the predictions.

Resumen En este trabajo el marco teórico, para desarrollar modelos genéricos basados en datos clínicos, se enfatiza e ilustra con estadísticas bayesianas las cuales predicen grados neurológicos para ilustrar los tipos de modelos que se pueden desarrollar desde un punto de vista matemático. Los modelos se basan en datos clínicos de pacientes que se han sometido a radioterapia por causa de un glioblastoma, el cual es un cáncer de cerebro agresivo. Un primer modelo requiere como parámetro el grado neurológico del paciente antes del tratamiento y predice el grado después del tratamiento. Un segundo modelo, mejorado, fue desarrollado con el propósito de hacerlo más real, éste emplea también el grado neurológico antes del tratamiento; además depende del Volumen Blanco Clínico (CTV por sus siglas en inglés). Por último, con el uso de estadísticas bayesianas fue posible estimar la incertidumbre de las predicciones.