A deep LSTM autoencoder-based framework for predictive maintenance of a proton radiotherapy delivery system.

INTRODUCTION: Unscheduled machine downtime can cause treatment interruptions and adversely impact patient treatment outcomes. Conventional Quality Assurance (QA) programs of a proton Pencil Beam Scanning (PBS) system ensure its operational performance by keeping the beam parameters within clinical tolerances but often do not reveal the underlying issues of the device prior to a machine malfunction event. In this study, we propose a Predictive Maintenance (PdM) approach that leverages an advanced analytical tool built on a deep neural network to detect treatment delivery machine issues early. METHODS: Beam delivery log file data from daily QA performed at the Burr Proton Center of Massachusetts General Hospital were collected. A novel PdM framework consisting of long short-term memory-based autoencoder (LSTM-AE) modeling of the proton PBS delivery system and a Mahalanobis distance-based error metric evaluation was constructed to detect rare anomalous machine events. These included QA beam pauses, clinical operational issues, and treatment interruptions. The model was trained in an unsupervised fashion on the QA data of normal sessions so that the model learned characteristics of normal machine operation. The anomaly is quantified as the multivariate deviation between the model predicted data and the measured data of the day using Mahalanobis distance (M-Score). Two-layer and three-layer Long short-term memory-based stacked autoencoder (LSTM-SAE) models were optimized for exploring model performance improvement. Model validation was performed with two clinical datasets and was analyzed using the area under the precision-recall curve (AUPRC) and the area under the receiver operating characteristic (AUROC). RESULTS: LSTM-SAE models showed strong performance in predicting QA beam pauses for both clinical validation datasets. Despite severe skew in the dataset, the model achieved AUPRC of 0.60 and 0.82 and AUROC of 0.75 and 0.92 in the respective 2018 and 2020 datasets. Moreover, these amount to 2.8-fold and 10.7-fold enhancement compared to the respective baseline event rates. In addition, in terms of treatment interruption events, model prediction enabled 3.88-fold and 51.2-fold detection improvement, while the detection improvement for clinical operational issues was 1.04-fold and 1.37-fold, respectively, in the 2018 and 2020 datasets. CONCLUSION: Our novel deep LSTM-SAE-based framework allows for highly discriminative prediction of anomalous machine events and demonstrates great promise for enabling PdM for proton PBS beam delivery.

Impacts of gantry angle dependent scanning beam properties on proton PBS treatment.

While proton beam models in treatment planning systems are generally assumed invariant with respect to the beam deliveries at different gantry angles. Physical properties of scanning pencil beams can change. The gantry angle dependent properties include the delivered charge to the monitor unit chamber, the spot position and the spot shape. The aim of this study is to investigate the extent of the changes and their dosimetric impacts using historical pencil beam scanning (PBS) treatment data. Online beam delivery records at the time of the patient-specific qualify assurance were retrospectively collected for a total of 34 PBS fields from 28 patients treated at our institution. For each field, proton beam properties at two different gantry angles (the planned and zero gantry angles) were extracted by a newly-developed machine log analysis method and used to reconstruct the delivered dose distributions in the cubic water phantom geometry. The reconstructed doses at the two different angles and a planar dose measurement by a 2D ion-chamber array were compared and the dosimetric impacts of the gantry angle dependency were accessed by a 3D γ-index analysis. In addition, the pencil beam spot size was independently characterized as a function of the gantry angle and the beam energy. The dosimetric effects of the perturbed beam shape were also investigated. Comparisons of spot-by-spot beam positions between both gantry angles show a mean deviation of 0.4 and 0.7 mm and a standard deviation of 0.3 and 0.6 mm for x and y directions, respectively. The delivered giga-protons per spot show a percent mean difference and a standard deviation of 0.01% and 0.3%, respectively, from each planned spot weight. These small deviations lead to an excellent agreement in dose comparisons with an average γ passing rate of 99.1%. When each calculation for both planned and zero gantry angles was compared to the measurement, a high correlation in γ values was also observed, also indicating the dosimetric differences are small when a field is delivered at different gantry angles. Utilizing the online beam delivery records, the gantry angle dependencies of the PBS beam delivery were assessed and quantified. The study confirms the variations of the physical properties to be sufficiently small within the clinical tolerances without taking into account the gantry angle variation.

Salivary gland cancer in BRCA-positive families: a retrospective review.

IMPORTANCE: Although an association between breast cancer and salivary gland cancers has been noted for decades, this is the first study, to our knowledge, to evaluate the possible linkage of BRCA gene mutations and this malignant neoplasm. OBJECTIVE: To compare the prevalence of salivary gland cancers in a large BRCA gene mutation database with background rates in the general population. DESIGN, SETTING, AND PARTICIPANTS: This is a retrospective review (June 1, 2012, through April 31, 2013) of pedigrees from patients with breast cancer in The Clinical Cancer Genetics Program at The Ohio State University Wexner Medical Center. A total of 5754 individuals were identified from 187 pedigrees, and their medical histories were reviewed for diagnoses of salivary gland tumors and BRCA testing. The pedigrees were restricted to provide a cohort of individuals with reasonable accuracy in family history by considering 3 generations of each pedigree, starting with the proband's generation and adding 1 generation above and below. The youngest generation was replaced with another older generation if there were no BRCA-related cancers or BRCA mutations recorded. Nonblood relatives of the proband (ie, stepparents and stepsiblings) were also excluded. MAIN OUTCOMES AND MEASURES: The rate of salivary gland cancers in the Clinical Center Genetics Program was compared with background incidence rates. RESULTS: After applying the restrictions to the 187 pedigrees in the database, 5754 individuals were included in the cohort. Two parotid gland cancers, 2 salivary gland cancers not otherwise specified, and 1 adenoid cystic carcinoma were identified. One of these cancers likely did not segregate with the BRCA mutation, and another individual tested negative for the BRCA mutation, resulting in a rate of 3 of 5754 (0.052%).The observed rate of 3 of 5754 cases (0.052%) of head and neck cancers in BRCA-positive probands and likely carriers is significantly higher than the background incidence rate of 3 of 100,000 (0.003%) per year (P < .001). CONCLUSIONS AND RELEVANCE: We believe this is a significant observation that, when considered alongside other similarities between salivary glands and breast tissue, warrants further investigation into the nature of a possible linkage between germline BRCA mutations and salivary gland cancer.