Automated, patient-specific estimation of regional imparted energy and dose from tube current modulated computed tomography exams across 13 protocols

Deep learning application engine (DLAE): Development and integration of deep learning algorithms in medical imaging

Herein we introduce a deep learning (DL) application engine (DLAE) system concept, present potential uses of it, and describe pathways for its integration in clinical workflows. An open-source software application was developed to provide a code-free approach to DL for medical imaging applications. DLAE supports several DL techniques used in medical imaging, including convolutional neural networks, fully convolutional networks, generative adversarial networks, and bounding box detectors. Several example applications using clinical images were developed and tested to demonstrate the capabilities of DLAE. Additionally, a model deployment example was demonstrated in which DLAE was used to integrate two trained models into a commercial clinical software package.

Toward automated and personalized organ dose determination in CT examinations - A comparison of two tissue characterization models for Monte Carlo organ dose calculation with a Therapy Planning System

PURPOSE

Computed tomography (CT) is a versatile tool in diagnostic radiology with rapidly increasing number of examinations per year globally. Routine adaption of the exposure level for patient anatomy and examination protocol cause the patients' exposures to become diversified and harder to predict by simple methods. To facilitate individualized organ dose estimates, we explore the possibility to automate organ dose calculations using a radiotherapy treatment planning system (TPS). In particular, the mapping of CT number to elemental composition for Monte Carlo (MC) dose calculations is investigated.

METHODS

Organ dose calculations were done for a female thorax examination test case with a TPS (Raystation™, Raysearch Laboratories AB, Stockholm, Sweden) utilizing a MC dose engine with a CT source model presented in a previous study. The TPS's inherent tissue characterization model for mapping of CT number to elemental composition of the tissues was calibrated using a phantom with known elemental compositions and validated through comparison of MC calculated dose with dose measured with Thermo Luminescence Dosimeters (TLD) in an anthropomorphic phantom. Given the segmentation tools of the TPS, organ segmentation strategies suitable for automation were analyzed for high contrast organs, utilizing CT number thresholding and model-based segmentation, and for low contrast organs utilizing water replacements in larger tissue volumes. Organ doses calculated with a selection of organ segmentation methods in combination with mapping of CT numbers to elemental composition (RT model), normally used in radiotherapy, were compared to a tissue characterization model with organ segmentation and elemental compositions defined by replacement materials [International Commission on Radiological Protection (ICRP) model], frequently favored in imaging dosimetry.

RESULTS

The results of the validation with the anthropomorphic phantom yielded mean deviations from the dose to water calculated with the RT and ICRP model as measured with TLD of 1.1% and 1.5% with maximum deviations of 6.1% and 8.7% respectively over all locations in the phantom. A strategy for automated organ segmentation was evaluated for two different risk organ groups, that is, low contrast soft organs and high contrast organs. The relative deviation between organ doses calculated with the RT model and with the ICRP model varied between 0% and 20% for the thorax/upper abdomen risk organs.

CONCLUSIONS

After calibration, the RT model in the TPS provides accurate MC dose results as compared to measurements with TLD and the ICRP model. Dosimetric feasible segmentation of the risk organs for a female thorax demonstrates a possibility for automation using the segmentation tool available in a TPS for high contrast organs. Low contrast soft organs can be represented by water volumes, but organ dose to the esophagus and thyroid must be determined using standardized organ shapes. The uncertainties of the organ doses are small compared to the overall uncertainty, at least an order of magnitude larger, in the estimates of lifetime attributable risk (LAR) based on organ doses. Large-scale and automated individual organ dose calculations could provide an improvement in cancer incidence estimates from epidemiological studies.