Anonymization of DICOM electronic medical records for radiation therapy

PRIMIS: Privacy-preserving medical image sharing via deep sparsifying transform learning with obfuscation

OBJECTIVE

The primary objective of our study is to address the challenge of confidentially sharing medical images across different centers. This is often a critical necessity in both clinical and research environments, yet restrictions typically exist due to privacy concerns. Our aim is to design a privacy-preserving data-sharing mechanism that allows medical images to be stored as encoded and obfuscated representations in the public domain without revealing any useful or recoverable content from the images. In tandem, we aim to provide authorized users with compact private keys that could be used to reconstruct the corresponding images.

METHOD

Our approach involves utilizing a neural auto-encoder. The convolutional filter outputs are passed through sparsifying transformations to produce multiple compact codes. Each code is responsible for reconstructing different attributes of the image. The key privacy-preserving element in this process is obfuscation through the use of specific pseudo-random noise. When applied to the codes, it becomes computationally infeasible for an attacker to guess the correct representation for all the codes, thereby preserving the privacy of the images.

RESULTS

The proposed framework was implemented and evaluated using chest X-ray images for different medical image analysis tasks, including classification, segmentation, and texture analysis. Additionally, we thoroughly assessed the robustness of our method against various attacks using both supervised and unsupervised algorithms.

CONCLUSION

This study provides a novel, optimized, and privacy-assured data-sharing mechanism for medical images, enabling multi-party sharing in a secure manner. While we have demonstrated its effectiveness with chest X-ray images, the mechanism can be utilized in other medical images modalities as well.

Anonymizing Radiographs Using an Object Detection Deep Learning Algorithm

Radiographic markers contain protected health information that must be removed before public release. This work presents a deep learning algorithm that localizes radiographic markers and selectively removes them to enable de-identified data sharing. The authors annotated 2000 hip and pelvic radiographs to train an object detection computer vision model. Data were split into training, validation, and test sets at the patient level. Extracted markers were then characterized using an image processing algorithm, and potentially useful markers (eg, "L" and "R") without identifying information were retained. The model achieved an area under the precision-recall curve of 0.96 on the internal test set. The de-identification accuracy was 100% (400 of 400), with a de-identification false-positive rate of 1% (eight of 632) and a retention accuracy of 93% (359 of 386) for laterality markers. The algorithm was further validated on an external dataset of chest radiographs, achieving a de-identification accuracy of 96% (221 of 231). After fine-tuning the model on 20 images from the external dataset to investigate the potential for improvement, a 99.6% (230 of 231, P = .04) de-identification accuracy and decreased false-positive rate of 5% (26 of 512) were achieved. These results demonstrate the effectiveness of a two-pass approach in image de-identification. Keywords: Conventional Radiography, Skeletal-Axial, Thorax, Experimental Investigations, Supervised Learning, Transfer Learning, Convolutional Neural Network (CNN) Supplemental material is available for this article. © RSNA, 2023 See also the commentary by Chang and Li in this issue.

Infrastructure platform for privacy-preserving distributed machine learning development of computer-assisted theragnostics in cancer

INTRODUCTION

Emerging evidence suggests that data-driven support tools have found their way into clinical decision-making in a number of areas, including cancer care. Improving them and widening their scope of availability in various differing clinical scenarios, including for prognostic models derived from retrospective data, requires co-ordinated data sharing between clinical centres, secondary analyses of large multi-institutional clinical trial data, or distributed (federated) learning infrastructures. A systematic approach to utilizing routinely collected data across cancer care clinics remains a significant challenge due to privacy, administrative and political barriers.

METHODS

An information technology infrastructure and web service software was developed and implemented which uses machine learning to construct clinical decision support systems in a privacy-preserving manner across datasets geographically distributed in different hospitals. The infrastructure was deployed in a network of Australian hospitals. A harmonized, international ontology-linked, set of lung cancer databases were built with the routine clinical and imaging data at each centre. The infrastructure was demonstrated with the development of logistic regression models to predict major cardiovascular events following radiation therapy.

RESULTS

The infrastructure implemented forms the basis of the Australian computer-assisted theragnostics (AusCAT) network for radiation oncology data extraction, reporting and distributed learning. Four radiation oncology departments (across seven hospitals) in New South Wales (NSW) participated in this demonstration study. Infrastructure was deployed at each centre and used to develop a model predicting for cardiovascular admission within a year of receiving curative radiotherapy for non-small cell lung cancer. A total of 10417 lung cancer patients were identified with 802 being eligible for the model. Twenty features were chosen for analysis from the clinical record and linked registries. After selection, 8 features were included and a logistic regression model achieved an area under the receiver operating characteristic (AUROC) curve of 0.70 and C-index of 0.65 on out-of-sample data.

CONCLUSION

The infrastructure developed was demonstrated to be usable in practice between clinical centres to harmonize routinely collected oncology data and develop models with federated learning. It provides a promising approach to enable further research studies in radiation oncology using real world clinical data.

A Two-Stage De-Identification Process for Privacy-Preserving Medical Image Analysis

Identification and re-identification are two major security and privacy threats to medical imaging data. De-identification in DICOM medical data is essential to preserve the privacy of patients' Personally Identifiable Information (PII) and requires a systematic approach. However, there is a lack of sufficient detail regarding the de-identification process of DICOM attributes, for example, what needs to be considered before removing a DICOM attribute. In this paper, we first highlight and review the key challenges in the medical image data de-identification process. In this paper, we develop a two-stage de-identification process for CT scan images available in DICOM file format. In the first stage of the de-identification process, the patient's PII-including name, date of birth, etc., are removed at the hospital facility using the export process available in their Picture Archiving and Communication System (PACS). The second stage employs the proposed DICOM de-identification tool for an exhaustive attribute-level investigation to further de-identify and ensure that all PII has been removed. Finally, we provide a roadmap for future considerations to build a semi-automated or automated tool for the DICOM datasets de-identification.

Automatic Curation of Court Documents: Anonymizing Personal Data

In order to provide open access to data of public interest, it is often necessary to perform several data curation processes. In some cases, such as biological databases, curation involves quality control to ensure reliable experimental support for biological sequence data. In others, such as medical records or judicial files, publication must not interfere with the right to privacy of the persons involved. There are also interventions in the published data with the aim of generating metadata that enable a better experience of querying and navigation. In all cases, the curation process constitutes a bottleneck that slows down general access to the data, so it is of great interest to have automatic or semi-automatic curation processes. In this paper, we present a solution aimed at the automatic curation of our National Jurisprudence Database, with special focus on the process of the anonymization of personal information. The anonymization process aims to hide the names of the participants involved in a lawsuit without losing the meaning of the narrative of facts. In order to achieve this goal, we need, not only to recognize person names but also resolve co-references in order to assign the same label to all mentions of the same person. Our corpus has significant differences in the spelling of person names, so it was clear from the beginning that pre-existing tools would not be able to reach a good performance. The challenge was to find a good way of injecting specialized knowledge about person names syntax while taking profit of previous capabilities of pre-trained tools. We fine-tuned an NER analyzer and we built a clusterization algorithm to solve co-references between named entities. We present our first results, which, for both tasks, are promising: We obtained a 90.21% of F1-micro in the NER task—from a 39.99% score before retraining the same analyzer in our corpus—and a 95.95% ARI score in clustering for co-reference resolution.

Segmentation of Organs and Tumor within Brain Magnetic Resonance Images Using K-Nearest Neighbor Classification

PURPOSE

To fully exploit the benefits of magnetic resonance imaging (MRI) for radiotherapy, it is desirable to develop segmentation methods to delineate patients' MRI images fast and accurately. The purpose of this work is to develop a semi-automatic method to segment organs and tumor within the brain on standard T1- and T2-weighted MRI images.

METHODS AND MATERIALS

Twelve brain cancer patients were retrospectively included in this study, and a simple rigid registration was used to align all the images to the same spatial coordinates. Regions of interest were created for organs and tumor segmentations. The K-nearest neighbor (KNN) classification algorithm was used to characterize the knowledge of previous segmentations using 15 image features (T1 and T2 image intensity, 4 Gabor filtered images, 6 image gradients, and 3 Cartesian coordinates), and the trained models were used to predict organ and tumor contours. Dice similarity coefficient (DSC), normalized surface dice, sensitivity, specificity, and Hausdorff distance were used to evaluate the performance of segmentations.

RESULTS

Our semi-automatic segmentations matched with the ground truths closely. The mean DSC value was between 0.49 (optical chiasm) and 0.89 (right eye) for organ segmentations and was 0.87 for tumor segmentation. Overall performance of our method is comparable or superior to the previous work, and the accuracy of our semi-automatic segmentation is generally better for large volume objects.

CONCLUSION

The proposed KNN method can accurately segment organs and tumor using standard brain MRI images, provides fast and accurate image processing and planning tools, and paves the way for clinical implementation of MRI-guided radiotherapy and adaptive radiotherapy.

A Pseudonymisation Protocol With Implicit and Explicit Consent Routes for Health Records in Federated Ledgers

Healthcare data for primary use (diagnosis) may be encrypted for confidentiality purposes; however, secondary uses such as feeding machine learning algorithms requires open access. Full anonymity has no traceable identifiers to report diagnosis results. Moreover, implicit and explicit consent routes are of practical importance under recent data protection regulations (GDPR), translating directly into break-the-glass requirements. Pseudonymisation is an acceptable compromise when dealing with such orthogonal requirements and is an advisable measure to protect data. Our work presents a pseudonymisation protocol that is compliant with implicit and explicit consent routes. The protocol is constructed on a (t,n)-threshold secret sharing scheme and public key cryptography. The pseudonym is safely derived from a fragment of public information without requiring any data-subject's secret. The method is proven secure under reasonable cryptographic assumptions and scalable from the experimental results.

Assessment of Differences in the Dimensions of Mandible Condyle Models in Fan- versus Cone-Beam Computer Tomography Acquisition

Modern treatment in the field of head and neck surgery aims for the least invasive therapy and places great emphasis on restorative treatment, especially in the case of injury and deformation corrective surgery. More and more often, surgeons use CAD/CAM (Computer-Aided Design and Computer-Aided Manufacturing) tools in their daily practice in the form of models, templates, and computer simulations of planning. These tools are based on DICOM (Digital Imaging and Communications in Medicine) files derived from computed tomography. They can be obtained from both fan-beam (FBCT) and cone-beam tomography (CBCT) acquisitions, which are subsequently segmented in order to transform them into a 1-bit 3D model, which is the basis for further CAD processes.

AIM

Evaluation of differences in the dimensions of mandible condyle models in fan- versus cone-beam computer tomography for surgical treatment purposes.

METHODS

499 healthy condyles were examined in CT-based 3D models of Caucasians aged 8-88 years old. Datasets were obtained from 66 CBCT and 184 FBCT axial image series (in each case, imaging both mandible condyles resulted in the acquisition of 132 condyles from CBCT and 368 condyles from FBCT) and were transformed into three-dimensional models by digital segmentation. Eleven different measurements were performed to obtain information whether there were any differences between FBCT and CBCT models of the same anatomical region.

RESULTS

7 of 11 dimensions were significantly higher in FBCT versus lower in CBCT (p < 0.05).

Proton Radiotherapy Could Reduce the Risk of Fatal Second Cancers for Children with Intracranial Tumors in Low- and Middle-Income Countries

PURPOSE

To test our hypothesis that, for young children with intracranial tumors, proton radiotherapy in a high-income country does not reduce the risk of a fatal subsequent malignant neoplasm (SMN) compared with photon radiotherapy in low- and middle-income countries.

MATERIALS AND METHODS

We retrospectively selected 9 pediatric patients with low-grade brain tumors who were treated with 3-dimensional conformal radiation therapy in low- and middle-income countries. Images and contours were deidentified and transferred to a high-income country proton therapy center. Clinically commissioned treatment planning systems of each academic hospital were used to calculate absorbed dose from the therapeutic fields. After fusing supplemental computational phantoms to the patients' anatomies, models from the literature were applied to calculate stray radiation doses. Equivalent doses were determined in organs and tissues at risk of SMNs, and the lifetime attributable risk of SMN mortality (LAR) was predicted using a dose-effect model. Our hypothesis test was based on the average of the ratios of LARs from proton therapy to that of photon therapy ()(H0: = 1; H A : < 1).

RESULTS

Proton therapy reduced the equivalent dose in organs at risk for SMNs and LARs compared with photon therapy for which the for the cohort was 0.69 ± 0.10, resulting in the rejection of H0 (P < .001, α = 0.05). We observed that the younger children in the cohort (2-4 years old) were at a factor of approximately 2.5 higher LAR compared with the older children (8-12 years old).

CONCLUSION

Our findings suggest that proton radiotherapy has the strong potential of reducing the risk of fatal SMNs in pediatric patients with intracranial tumors if it were made available globally.

Clinico-anatomical classification of the processus condylaris mandibulae for traumatological purposes

Mandible condyle fracture has been reported to constitute 9-45 % (Asprino and Consani, 2006), 14.1 % (Bataineh, 1998), 25-50 % (Silvennoinen, Iizuka, 1992), 32 % (Chrcanovic et al., 2004), and 38 % (Brasileiro and Passeri, 2006) of all mandible fractures (Kozakiewicz and Swiniarski, 2013). Small bone segments, limited available space for application of the fixation material and limited visibility of the operative field are common difficulties. To guarantee satisfactory treatment effects, anatomical reduction and proper fracture stability are necessary. The use of 3-4 screws in the upper section (proximal segment) provides adequate immobilization, which can be easily achieved when the condyle is low and wide. However, if the condyle is slender, it is not technically possible to fix 2 plates and 4 screws for osteosynthesis. Selection of the appropriate fixative material that will provide adequate rigidity during the healing period while simultaneously allowing proper construction of the lateral silhouette of the processus condylaris mandibulae to fix the plate remains a key consideration. The aim of this study was to evaluate clinico-anatomical classification of the condyle of mandible posture for traumatological purposes. Five hundred computer tomography virtual models were created, from which 11 measurements were made, and 2 indexes were calculated. Assessment of types based on the ratio of the condyle height index revealed a dichotomous division into high and short condyles. Statistically associated with the division, the Width_neck_basal (the width of the bone at the level of semilunar notch measured by a frontal projection perpendicular to line "A", as described by Neff (Neff et al., 2014)) measurement allowed the creation of the following clinico-anatomical classification: -slender-type condyles have a Width_neck_basal in the range of 4-8.5mm; -squad-type condyles have a Width_neck_basal in the range of 11.5-19.5mm. Patients with a Width_neck_basal value in the 8.5-11.5mm range cannot be classified using this method, and a different method to assess the lateral condylar silhouette must be used. The proposed clinic-anatomical classification method avoids the problems associated with incorrect osteosynthesis plate selection. Assignment to a group can be obtained by making one measurement (the Width_neck_basal). In that way, the optimal fixing material can be selected by the surgeon before the operation commences, with great intraoperation time savings.

Comparison of conventional and advanced radiotherapy techniques for left-sided breast cancer after breast conserving surgery

Whole breast radiotherapy (WBRT) after breast conserving surgery is the standard treatment to prevent recurrence and metastasis of early stage breast cancer. This study aims to compare seven WBRT techniques including conventional tangential, field-in-field (FIF), hybrid intensity-modulated radiotherapy (IMRT), IMRT, standard volumetric modulated arc therapy (STD-VMAT), noncoplanar VMAT (NC-VMAT), and multiple arc VMAT (MA-VMAT). Fifteen patients who were previously diagnosed with left-sided early stage breast cancer and treated in our clinic were selected for this study. WBRT plans were created for these patients and were evaluated based on target coverage and normal tissue toxicities. All techniques produced clinically acceptable WBRT plans. STD-VMAT delivered the lowest mean dose (1.1 ± 0.3 Gy) and the lowest maximum dose (7.3 ± 4.9 Gy) to contralateral breast, and the second lowest lifetime attributable risk (LAR) (4.1 ± 1.4%) of secondary contralateral breast cancer. MA-VMAT delivered the lowest mean dose to lungs (4.9 ± 0.9 Gy) and heart (5.5 ± 1.2 Gy), exhibited the lowest LAR (1.7 ± 0.3%) of secondary lung cancer, normal tissue complication probability (NTCP) (1.2 ± 0.2%) of pneumonitis, risk of coronary events (RCE) (10.3 ± 2.7%), and LAR (3.9 ± 1.3%) of secondary contralateral breast cancer. NC-VMAT plans provided the most conformal target coverage, the lowest maximum lung dose (46.2 ± 4.1 Gy) and heart dose (41.1 ± 5.4 Gy), and the second lowest LAR (1.8 ± 0.4%) of secondary lung cancer and RCE (10.5 ± 2.8%). MA-VMAT and NC-VMAT could be the preferred techniques for early stage breast cancer patients after breast conserving surgery.

De-Identification of Radiomics Data Retaining Longitudinal Temporal Information

We propose a de-identification system which runs in a standalone mode. The system takes care of the de-identification of radiation oncology patient's clinical and annotated imaging data including RTSTRUCT, RTPLAN, and RTDOSE. The clinical data consists of diagnosis, stages, outcome, and treatment information of the patient. The imaging data could be the diagnostic, therapy planning, and verification images. Archival of the longitudinal radiation oncology verification images like cone beam CT scans along with the initial imaging and clinical data are preserved in the process. During the de-identification, the system keeps the reference of original data identity in encrypted form. These could be useful for the re-identification if necessary.

Linear, planar and spherical tensor-valued diffusion MRI data by free waveform encoding in healthy brain, water, oil and liquid crystals

Recently, several biophysical models and signal representations have been proposed for microstructure imaging based on tensor-valued, or multidimensional, diffusion MRI. The acquisition of the necessary data requires non-conventional pulse sequences, and data is therefore not available to the wider diffusion MRI community. To facilitate exploration and development of analysis techniques based on tensor-valued diffusion encoding, we share a comprehensive data set acquired in a healthy human brain. The data encompasses diffusion weighted images using linear, planar and spherical diffusion tensor encoding at multiple b-values and diffusion encoding directions. We also supply data acquired in several phantoms that may support validation. The data is hosted by GitHub: https://github.com/filip-szczepankiewicz/Szczepankiewicz_DIB_2019.

Postmastectomy radiotherapy for left-sided breast cancer patients: Comparison of advanced techniques

Postmastectomy radiotherapy (PMRT) has been shown to improve the overall survival for invasive breast cancer patients, and many advanced radiotherapy technologies were adopted for PMRT. The purpose of our study is to compare various advanced PMRT techniques including fixed-beam intensity-modulated radiotherapy (IMRT), non-coplanar volumetric modulated arc therapy (NC-VMAT), multiple arc VMAT (MA-VMAT), and tomotherapy (TOMO). Results of standard VMAT and mixed beam therapy that were published by our group previously were also included in the plan comparisons. Treatment plans were produced for nine PMRT patients previously treated in our clinic. The plans were evaluated based on planning target volume (PTV) coverage, dose homogeneity index (DHI), conformity index (CI), dose to organs at risk (OARs), normal tissue complication probability (NTCP) of pneumonitis, lifetime attributable risk (LAR) of second cancers, and risk of coronary events (RCE). All techniques produced clinically acceptable PMRT plans. Overall, fixed-beam IMRT delivered the lowest mean dose to contralateral breast (1.56 ± 0.4 Gy) and exhibited lowest LAR (0.6 ± 0.2%) of secondary contralateral breast cancer; NC-VMAT delivered the lowest mean dose to lungs (7.5 ± 0.8 Gy), exhibited lowest LAR (5.4 ± 2.8%) of secondary lung cancer and lowest NTCP (2.1 ± 0.4%) of pneumonitis; mixed beam therapy delivered the lowest mean dose to heart (7.1 ± 1.3 Gy) and exhibited lowest RCE (8.6 ± 7.1%); TOMO plans provided the most optimal target coverage while delivering higher dose to OARs than other techniques. Both NC-VMAT and MA-VMAT exhibited lower values of all OARs evaluation metrics compare to standard VMAT. Fixed-beam IMRT, NC-VMAT, and mixed beam therapy could be the optimal radiation technique for certain breast cancer patients after mastectomy.

An atlas-based method to predict three-dimensional dose distributions for cancer patients who receive radiotherapy

Due to the complexity of advanced radiotherapy techniques, treatment planning process is usually time consuming and plan quality can vary considerably among planners and institutions. It is also impractical to generate all possible treatment plans based on available radiotherapy techniques and select the best option for a specific patient. Automatic dose prediction will be very helpful in these situations, while there were a few studies of three-dimensional (3D) dose prediction for patients who received radiotherapy. The purpose of this work was to develop a novel atlas-based method to predict 3D dose prediction and to evaluate its performance. Previously treated nineteen left-sided post-mastectomy breast cancer patients and sixteen prostate cancer patients were included in this study. One patient was arbitrarily chosen as the reference for each type of cancer and all the remaining patients' computed tomography (CT) images and contours were aligned to it using deformable image registration (DIR). Deformable vector field (DVF) for each patient i (DVF _i-ref) was used to deform the original 3D dose matrix of that patient. CT scan of a test patient was also registered with the same reference patient using DIR and both direct DVF (DVF_test-ref) and inverse DVF ([Formula: see text]) were derived. Similarity of atlas patients to the test patient was determined based on the similarity of DVF_test-ref to atlas DVFs (DVF _i-ref) and appropriate weighting factors were calculated. Patients' doses in the atlas were deformed again using [Formula: see text] to transform them from the reference patient's coordinates to the test patient's coordinates and the final 3D dose distribution for the test patient was predicted by summing the weighted individual 3D dose distributions. Performance of our method was evaluated and the results revealed that the proposed method was able to predict the 3D dose distributions accurately. The mean dose difference between clinical and predicted 3D dose distributions were 0.9  ±  1.1 Gy and 1.9  ±  1.2 Gy for breast and prostate plans. The proposed dose prediction method can be used to improve planning quality and facilitate plan comparisons.

Identification and classification of DICOM files with burned-in text content

BACKGROUND

Protected health information burned in pixel data is not indicated for various reasons in DICOM. It complicates the secondary use of such data. In recent years, there have been several attempts to anonymize or de-identify DICOM files. Existing approaches have different constraints. No completely reliable solution exists. Especially for large datasets, it is necessary to quickly analyse and identify files potentially violating privacy.

METHODS

Classification is based on adaptive-iterative algorithm designed to identify one of three classes. There are several image transformations, optical character recognition, and filters; then a local decision is made. A confirmed local decision is the final one. The classifier was trained on a dataset composed of 15,334 images of various modalities.

RESULTS

The false positive rates are in all cases below 4.00%, and 1.81% in the mission-critical problem of detecting protected health information. The classifier's weighted average recall was 94.85%, the weighted average inverse recall was 97.42% and Cohen's Kappa coefficient was 0.920.

CONCLUSION

The proposed novel approach for classification of burned-in text is highly configurable and able to analyse images from different modalities with a noisy background. The solution was validated and is intended to identify DICOM files that need to have restricted access or be thoroughly de-identified due to privacy issues. Unlike with existing tools, the recognised text, including its coordinates, can be further used for de-identification.

Low- and middle-income countries can reduce risks of subsequent neoplasms by referring pediatric craniospinal cases to centralized proton treatment centers

Few children with cancer in low- and middle-income countries (LMICs) have access to proton therapy. Evidence exists to support replacing photon therapy with proton therapy to reduce the incidence of secondary malignant neoplasms (SMNs) in childhood cancer survivors. The purpose of this study was to estimate the potential reduction in SMN incidence and in SMN mortality for pediatric medulloblastoma patients in LMICs if proton therapy were made available to them. For nine children of ages 2 to 14 years, we calculated the equivalent dose in organs or tissues at risk for radiogenic SMNs from therapeutic and stray radiation for photon craniospinal irradiation (CSI) in a LMIC and proton CSI in a high-income country. We projected the lifetime risks of SMN incidence and SMN mortality for every SMN site with a widely-used model from the literature. We found that the average total lifetime attributable risks of incidence and mortality were very high for both photon CSI (168% and 41%, respectively) and proton CSI (88% and 26%, respectively). SMNs having the highest risk of mortality were lung cancer (16%), non-site-specific solid tumors (16%), colon cancer (5.9%), leukemia (5.4%), and for girls breast cancer (5.0%) after photon CSI and non-site-specific solid tumors (12%), lung cancer (11%), and leukemia (4.8%) after proton CSI. The risks were higher for younger children than for older children and higher for girls than for boys. The ratios of proton CSI to photon CSI of total risks of SMN incidence and mortality were 0.56 (95% CI, 0.37 to 0.75) and 0.64 (95% CI, 0.45 to 0.82), respectively, averaged over this sample group. In conclusion, proton therapy has the potential to lessen markedly subsequent SMNs and SMN fatalities in survivors of childhood medulloblastoma in LMICs, for example, through regional centralized care. Additional methods should be explored urgently to reduce therapeutic-field doses in organs and tissues at risk for SMN, especially in the lungs, colon, and breast tissues.

Supplemental computational phantoms to estimate out-of-field absorbed dose in photon radiotherapy

The purpose of this study was to develop a straightforward method of supplementing patient anatomy and estimating out-of-field absorbed dose for a cohort of pediatric radiotherapy patients with limited recorded anatomy. A cohort of nine children, aged 2-14 years, who received 3D conformal radiotherapy for low-grade localized brain tumors (LBTs), were randomly selected for this study. The extent of these patients' computed tomography simulation image sets were cranial only. To approximate their missing anatomy, we supplemented the LBT patients' image sets with computed tomography images of patients in a previous study with larger extents of matched sex, height, and mass and for whom contours of organs at risk for radiogenic cancer had already been delineated. Rigid fusion was performed between the LBT patients' data and that of the supplemental computational phantoms using commercial software and in-house codes. In-field dose was calculated with a clinically commissioned treatment planning system, and out-of-field dose was estimated with a previously developed analytical model that was re-fit with parameters based on new measurements for intracranial radiotherapy. Mean doses greater than 1 Gy were found in the red bone marrow, remainder, thyroid, and skin of the patients in this study. Mean organ doses between 150 mGy and 1 Gy were observed in the breast tissue of the girls and lungs of all patients. Distant organs, i.e. prostate, bladder, uterus, and colon, received mean organ doses less than 150 mGy. The mean organ doses of the younger, smaller LBT patients (0-4 years old) were a factor of 2.4 greater than those of the older, larger patients (8-12 years old). Our findings demonstrated the feasibility of a straightforward method of applying supplemental computational phantoms and dose-calculation models to estimate absorbed dose for a set of children of various ages who received radiotherapy and for whom anatomies were largely missing in their original computed tomography simulations.

Controlled searching in reversibly de-identified medical imaging archives

Nowadays, digital medical imaging in healthcare has become a fundamental tool for medical diagnosis. This growth has been accompanied by the development of technologies and standards, such as the DICOM standard and PACS. This environment led to the creation of collaborative projects where there is a need to share medical data between different institutions for research and educational purposes. In this context, it is necessary to maintain patient data privacy and provide an easy and secure mechanism for authorized personnel access. This paper presents a solution that fully de-identifies standard medical imaging objects, including metadata and pixel data, providing at the same time a reversible de-identifier mechanism that retains search capabilities from the original data. The last feature is important in some scenarios, for instance, in collaborative platforms where data is anonymized when shared with the community but searchable for data custodians or authorized entities. The solution was integrated into an open source PACS archive and validated in a multidisciplinary collaborative scenario.

Matched computed tomography segmentation and demographic data for oropharyngeal cancer radiomics challenges

Cancers arising from the oropharynx have become increasingly more studied in the past few years, as they are now epidemic domestically. These tumors are treated with definitive (chemo)radiotherapy, and have local recurrence as a primary mode of clinical failure. Recent data suggest that ‘radiomics’, or extraction of image texture analysis to generate mineable quantitative data from medical images, can reflect phenotypes for various cancers. Several groups have shown that developed radiomic signatures, in head and neck cancers, can be correlated with survival outcomes. This data descriptor defines a repository for head and neck radiomic challenges, executed via a Kaggle in Class platform, in partnership with the MICCAI society 2016 annual meeting.These public challenges were designed to leverage radiomics and/or machine learning workflows to discriminate HPV phenotype in one challenge (HPV status challenge) and to identify patients who will develop a local recurrence in the primary tumor volume in the second one (Local recurrence prediction challenge) in a segmented, clinically curated anonymized oropharyngeal cancer (OPC) data set.

A De-Identification Pipeline for Ultrasound Medical Images in DICOM Format

Clinical data sharing between healthcare institutions, and between practitioners is often hindered by privacy protection requirements. This problem is critical in collaborative scenarios where data sharing is fundamental for establishing a workflow among parties. The anonymization of patient information burned in DICOM images requires elaborate processes somewhat more complex than simple de-identification of textual information. Usually, before sharing, there is a need for manual removal of specific areas containing sensitive information in the images. In this paper, we present a pipeline for ultrasound medical image de-identification, provided as a free anonymization REST service for medical image applications, and a Software-as-a-Service to streamline automatic de-identification of medical images, which is freely available for end-users. The proposed approach applies image processing functions and machine-learning models to bring about an automatic system to anonymize medical images. To perform character recognition, we evaluated several machine-learning models, being Convolutional Neural Networks (CNN) selected as the best approach. For accessing the system quality, 500 processed images were manually inspected showing an anonymization rate of 89.2%. The tool can be accessed at https://bioinformatics.ua.pt/dicom/anonymizer and it is available with the most recent version of Google Chrome, Mozilla Firefox and Safari. A Docker image containing the proposed service is also publicly available for the community.

A Review of Radiotherapy-Induced Late Effects Research after Advanced Technology Treatments

The number of incident cancers and long-term cancer survivors is expected to increase substantially for at least a decade. Advanced technology radiotherapies, e.g., using beams of protons and photons, offer dosimetric advantages that theoretically yield better outcomes. In general, evidence from controlled clinical trials and epidemiology studies are lacking. To conduct these studies, new research methods and infrastructure will be needed. In the paper, we review several key research methods of relevance to late effects after advanced technology proton-beam and photon-beam radiotherapies. In particular, we focus on the determination of exposures to therapeutic and stray radiation and related uncertainties, with discussion of recent advances in exposure calculation methods, uncertainties, in silico studies, computing infrastructure, electronic medical records, and risk visualization. We identify six key areas of methodology and infrastructure that will be needed to conduct future outcome studies of radiation late effects.

A machine learning methodology for medical imaging anonymization

Privacy protection is a major requirement for the complete success of EHR systems, becoming even more critical in collaborative scenarios, where data is shared among institutions and practitioners. While textual data can be easily de-identified, patient data in medical images implies a more elaborate approach. In this work we present a solution for sensitive word identification in medical images based on a combination of two machine-learning models, achieving a F1-score of 0.94. Three experts evaluated the system performance. They analyzed the output of the present methodology and categorized the studies in three groups: studies that had their sensitive words removed (true positive), studies with complete patient identity (false negative) and studies with mistakenly removed data (false positive). The experts were unanimous regarding the relevance of the present tool in collaborative medical environments, as it may improve the exchange of anonymized patient data between institutions.

Visualization of risk of radiogenic second cancer in the organs and tissues of the human body

BACKGROUND

Radiogenic second cancer is a common late effect in long term cancer survivors. Currently there are few methods or tools available to visually evaluate the spatial distribution of risks of radiogenic late effects in the human body. We developed a risk visualization method and demonstrated it for radiogenic second cancers in tissues and organs of one patient treated with photon volumetric modulated arc therapy and one patient treated with proton craniospinal irradiation.

METHODS

Treatment plans were generated using radiotherapy treatment planning systems (TPS) and dose information was obtained from TPS. Linear non-threshold risk coefficients for organs at risk of second cancer incidence were taken from the Biological Effects of Ionization Radiation VII report. Alternative risk models including linear exponential model and linear plateau model were also examined. The predicted absolute lifetime risk distributions were visualized together with images of the patient anatomy.

RESULTS

The risk distributions of second cancer for the two patients were visually presented. The risk distributions varied with tissue, dose, dose-risk model used, and the risk distribution could be similar to or very different from the dose distribution.

CONCLUSIONS

Our method provides a convenient way to directly visualize and evaluate the risks of radiogenic second cancer in organs and tissues of the human body. In the future, visual assessment of risk distribution could be an influential determinant for treatment plan scoring.

Reducing the cost of proton radiation therapy: the feasibility of a streamlined treatment technique for prostate cancer

Proton radiation therapy is an effective modality for cancer treatments, but the cost of proton therapy is much higher compared to conventional radiotherapy and this presents a formidable barrier to most clinical practices that wish to offer proton therapy. Little attention in literature has been paid to the costs associated with collimators, range compensators and hypofractionation. The objective of this study was to evaluate the feasibility of cost-saving modifications to the present standard of care for proton treatments for prostate cancer. In particular, we quantified the dosimetric impact of a treatment technique in which custom fabricated collimators were replaced with a multileaf collimator (MLC) and the custom range compensators (RC) were eliminated. The dosimetric impacts of these modifications were assessed for 10 patients with a commercial treatment planning system (TPS) and confirmed with corresponding Monte Carlo simulations. We assessed the impact on lifetime risks of radiogenic second cancers using detailed dose reconstructions and predictive dose-risk models based on epidemiologic data. We also performed illustrative calculations, using an isoeffect model, to examine the potential for hypofractionation. Specifically, we bracketed plausible intervals of proton fraction size and total treatment dose that were equivalent to a conventional photon treatment of 79.2 Gy in 44 fractions. Our results revealed that eliminating the RC and using an MLC had negligible effect on predicted dose distributions and second cancer risks. Even modest hypofractionation strategies can yield substantial cost savings. Together, our results suggest that it is feasible to modify the standard of care to increase treatment efficiency, reduce treatment costs to patients and insurers, while preserving high treatment quality.

Implementation of an analytical model for leakage neutron equivalent dose in a proton radiotherapy planning system

Equivalent dose from neutrons produced during proton radiotherapy increases the predicted risk of radiogenic late effects. However, out-of-field neutron dose is not taken into account by commercial proton radiotherapy treatment planning systems. The purpose of this study was to demonstrate the feasibility of implementing an analytical model to calculate leakage neutron equivalent dose in a treatment planning system. Passive scattering proton treatment plans were created for a water phantom and for a patient. For both the phantom and patient, the neutron equivalent doses were small but non-negligible and extended far beyond the therapeutic field. The time required for neutron equivalent dose calculation was 1.6 times longer than that required for proton dose calculation, with a total calculation time of less than 1 h on one processor for both treatment plans. Our results demonstrate that it is feasible to predict neutron equivalent dose distributions using an analytical dose algorithm for individual patients with irregular surfaces and internal tissue heterogeneities. Eventually, personalized estimates of neutron equivalent dose to organs far from the treatment field may guide clinicians to create treatment plans that reduce the risk of late effects.

Inter-Institutional Comparison of Personalized Risk Assessments for Second Malignant Neoplasms for a 13-Year-Old Girl Receiving Proton versus Photon Craniospinal Irradiation

Children receiving radiotherapy face the probability of a subsequent malignant neoplasm (SMN). In some cases, the predicted SMN risk can be reduced by proton therapy. The purpose of this study was to apply the most comprehensive dose assessment methods to estimate the reduction in SMN risk after proton therapy vs. photon therapy for a 13-year-old girl requiring craniospinal irradiation (CSI). We reconstructed the equivalent dose throughout the patient’s body from therapeutic and stray radiation and applied SMN incidence and mortality risk models for each modality. Excluding skin cancer, the risk of incidence after proton CSI was a third of that of photon CSI. The predicted absolute SMN risks were high. For photon CSI, the SMN incidence rates greater than 10% were for thyroid, non-melanoma skin, lung, colon, stomach, and other solid cancers, and for proton CSI they were non-melanoma skin, lung, and other solid cancers. In each setting, lung cancer accounted for half the risk of mortality. In conclusion, the predicted SMN risk for a 13-year-old girl undergoing proton CSI was reduced vs. photon CSI. This study demonstrates the feasibility of inter-institutional whole-body dose and risk assessments and also serves as a model for including risk estimation in personalized cancer care.