Search NASA⌕ Search

SEARCH · Search NASA

Results for “Cancer models”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 451 records · Page 25

Antibiotic drug tigecycline inhibited cell proliferation and induced autophagy in gastric cancer cells

Highlights: • Tigecycline inhibited cell growth and proliferation in human gastric cancer cells. • Tigecycline induced autophagy not apoptosis in human gastric cancer cells. • AMPK/mTOR/p70S6K pathway was activated after tigecycline treatment. • Tigecycline inhibited tumor growth in xenograft model of human gastric cancer cells. - Abstract: Tigecycline acts as a glycylcycline class bacteriostatic agent, and actively resists a series of bacteria, specifically drug fast bacteria. However, accumulating evidence showed that tetracycline and their derivatives such as doxycycline and minocycline have anti-cancer properties, which are out of their broader antimicrobial activity. We found that tigecycline dramatically inhibited gastric cancer cell proliferation and provided an evidence that tigecycline induced autophagy but not apoptosis in human gastric cancer cells. Further experiments demonstrated that AMPK pathway was activated accompanied with the suppression of its downstream targets including mTOR and p70S6K, and ultimately induced cell autophagy and inhibited cell growth. So our data suggested that tigecycline might act as a candidate agent for pre-clinical evaluation in treatment of patients suffering from gastric cancer.

60 APPLIED LIFE SCIENCES↗

SU-E-J-244: Development and Validation of a Knowledge Based Planning Model for External Beam Radiation Therapy of Locally Advanced Non-Small Cell Lung Cancer

Purpose: The study aims to develop and validate a knowledge based planning (KBP) model for external beam radiation therapy of locally advanced non-small cell lung cancer (LA-NSCLC). Methods: RapidPlan™ technology was used to develop a lung KBP model. Plans from 65 patients with LA-NSCLC were used to train the model. 25 patients were treated with VMAT, and the other patients were treated with IMRT. Organs-at-risk (OARs) included right lung, left lung, heart, esophagus, and spinal cord. DVH and geometric distribution DVH were extracted from the treated plans. The model was trained using principal component analysis and step-wise multiple regression. Box plot and regression plot tools were used to identify geometric outliers and dosimetry outliers and help fine-tune the model. The validation was performed by (a) comparing predicted DVH boundaries to actual DVHs of 63 patients and (b) using an independent set of treatment planning data. Results: 63 out of 65 plans were included in the final KBP model with PTV volume ranging from 102.5cc to 1450.2cc. Total treatment dose prescription varied from 50Gy to 70Gy based on institutional guidelines. One patient was excluded due to geometric outlier where 2.18cc of spinal cord was included in PTV. The other patient was excluded due to dosimetric outlier where the dose sparing to spinal cord was heavily enforced in the clinical plan. Target volume, OAR volume, OAR overlap volume percentage to target, and OAR out-of-field volume were included in the trained model. Lungs and heart had two principal component scores of GEDVH, whereas spinal cord and esophagus had three in the final model. Predicted DVH band (mean ±1 standard deviation) represented 66.2±3.6% of all DVHs. Conclusion: A KBP model was developed and validated for radiotherapy of LA-NSCLC in a commercial treatment planning system. The clinical implementation may improve the consistency of IMRT/VMAT planning.

60 APPLIED LIFE SCIENCES↗

In with the old, in with the new: machine learning for time to event biomedical research

The predictive modeling literature for biomedical applications is dominated by biostatistical methods for survival analysis, and more recently some out of the box machine learning approaches. In this article, we show a presentation of a machine learning method appropriate for time-to-event modeling in the area of prostate cancer long-term disease progression. Using XGBoost adapted to long-term disease progression, we developed a predictive model for 118 788 patients with localized prostate cancer at diagnosis from the Department of Veterans Affairs (VA). Our model accounted for patient censoring. Harrell’s c-index for our model using only features available at the time of diagnosis was 0.757 95% confidence interval [0.756, 0.757]. Our results show that machine learning methods like XGBoost can be adapted to use accelerated failure time (AFT) with censoring to model long-term risk of disease progression. Furthermore, the long median survival justifies and requires censoring. Overall, we show that an existing machine learning approach can be used for AFT outcome modeling in prostate cancer, and more generally for other chronic diseases with long observation times.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗

The TLR4 agonist fibronectin extra domain A is cryptic, exposed by elastase-2; use in a fibrin matrix cancer vaccine

Fibronectin (FN) is an extracellular matrix (ECM) protein including numerous fibronectin type III (FNIII) repeats with different functions. The alternatively spliced FN variant containing the extra domain A (FNIII EDA), located between FNIII 11 and FNIII 12, is expressed in sites of injury, chronic inflammation, and solid tumors. Although its function is not well understood, FNIII EDA is known to agonize Toll-like receptor 4 (TLR4). Here, by producing various FN fragments containing FNIII EDA, we found that FNIII EDA's immunological activity depends upon its local intramolecular context within the FN chain. N-terminal extension of the isolated FNIII EDA with its neighboring FNIII repeats (FNIII 9-10-11) enhanced its activity in agonizing TLR4, while C-terminal extension with the native FNIII 12-13-14 heparin-binding domain abrogated it. We reveal that an elastase 2 cleavage site is present between FNIII EDA and FNIII 12. Activity of the C-terminally extended FNIII EDA could be restored after cleavage of the FNIII 12-13-14 domain by elastase 2. FN being naturally bound to the ECM, we immobilized FNIII EDA-containing FN fragments within a fibrin matrix model along with antigenic peptides. Such matrices were shown to stimulate cytotoxic CD8 + T cell responses in two murine cancer models.

59 BASIC BIOLOGICAL SCIENCES↗

SU-F-T-25: Design and Implementation of a Multi-Purpose Applicator for Pelvic Brachytherapy Applications

Purpose: The current generation of inflatable multichannel brachytherapy applicators, such as the Varian Capri, have limited implementation to only vaginal and rectal cancers. While there are similar designs utilizing rigid, non-inflatable applicators, these alternatives could cause increased dose to surrounding tissue due to air gaps. Modification of the Capri could allow for easier treatment planning by reducing the number of channels and increased versatility by modifying the applicator to include an attachable single tandem for cervical or multiple tandems for endometrial applications. Methods: A Varian Capri applicator was simulated in water to replicate a patient. Multiple plans were optimized to deliver a prescribed dose of 100 cGy at 5mm away from the exterior of the applicator using six to thirteen existing channels. The current model was expanded upon to include a detachable tandem or multiple tandoms to increase its functionality to both cervical and endometrial cancers. Models were constructed in both threedimensional rendering software and Monte Carlo to allow prototyping and simulations. Results: Treatment plans utilizing six to thirteen channels produced limited dosimetric differences between channel arrangements, with a seven channel plan very closely approximating the thirteen channels. It was concluded that only seven channels would be necessary in future simulations to give an accurate representation of the applicator. Tandem attachments were prototyped for the applicator to demonstrate the ease of which they could be included. Future simulation in treatment planning software and Monte Carlo results will be presented to further define the ideal applicator geometry Conclusion: The current Capri applicator design could be easily modified to increase applicability to include cervical and endometrial treatments in addition to vaginal and rectal cancers. This new design helps in a more versatile single use applicator that can easily be inserted and to further reduce dose to critical structures during brachytherapy treatments.

60 APPLIED LIFE SCIENCES↗

A new view of radiation-induced cancer: integrating short- and long-term processes. Part II: second cancer risk estimation

As the number of cancer survivors grows, prediction of radiotherapy-induced second cancer risks becomes increasingly important. Because the latency period for solid tumors is long, the risks of recently introduced radiotherapy protocols are not yet directly measurable. In the accompanying article, we presented a new biologically based mathematical model, which, in principle, can estimate second cancer risks for any protocol. The novelty of the model is that it integrates, into a single formalism, mechanistic analyses of pre-malignant cell dynamics on two different time scales: short-term during radiotherapy and recovery; long-term during the entire life span. Here, we apply the model to nine solid cancer types (stomach, lung, colon, rectal, pancreatic, bladder, breast, central nervous system, and thyroid) using data on radiotherapyinduced second malignancies, on Japanese atomic bomb survivors, and on background US cancer incidence. Potentially, the model can be incorporated into radiotherapy treatment planning algorithms, adding second cancer risk as an optimization criterion.

59 BASIC BIOLOGICAL SCIENCES↗

Normal Tissue Complication Probability (NTCP) Prediction Model for Osteoradionecrosis of the Mandible in Patients With Head and Neck Cancer After Radiation Therapy: Large-Scale Observational Cohort

Osteoradionecrosis (ORN) of the mandible represents a severe, debilitating complication of radiation therapy (RT) for head and neck cancer (HNC). At present, no normal tissue complication probability (NTCP) models for risk of ORN exist. The aim of this study was to develop a multivariable clinical/dose-based NTCP model for the prediction of ORN any grade (ORN{sub I-IV}) and grade IV (ORN{sub IV}) after RT (±chemotherapy) in patients with HNC.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Comparison and Limitations of DVH-Based NTCP Models Derived From 3D-CRT and IMRT Data for Prediction of Gastrointestinal Toxicities in Prostate Cancer Patients by Using Propensity Score Matched Pair Analysis

Purpose: This study compared normal tissue complication probability (NTCP) modeling of chronic gastrointestinal toxicities following prostate cancer treatment for 2 treatment modalities. Possible factors causing discrepancies in optimal NTCP model parameters between 3-dimensional conformal radiation therapy (3D-CRT) and intensity modulated RT (IMRT) were analyzed and discussed, including the impact of patient characteristics, image guidance, toxicity scoring bias, and NTCP model limitations. Methods and Materials: Rectal wall dose-volume histograms of 1115 patients treated for prostate cancer under an adaptive radiation therapy protocol were used to model gastrointestinal toxicity grade ≥2 (according to Common Terminology Criteria for Adverse Events). A total of 457 patients were treated with 3D-CRT and 658 with IMRT. 3D-CRT patients were matched to IMRT patients based on various patient characteristics, using a propensity score–based algorithm. Parameters of the Lyman equivalent uniform dose and cut-off dose logistic regression NTCP models were estimated for the 2 matched treatment modalities and the combined group. Results: After they were matched, the 3D-CRT and IMRT groups contained 275 and 550 patients with a large discrepancy of 28.7% versus 7.8% toxicities, respectively (P<.001). For both NTCP models, optimal parameters found for the 3D-CRT groups did not fit the IMRT patients well and vice versa. Models developed for the combined data overestimated NTCP for the IMRT patients and underestimated NTCP for the 3D-CRT group. Conclusions: Our analysis did not reveal a single definitive cause for discrepancies of model parameters between 3D-CRT and IMRT. Patient characteristics and bias in toxicity scoring, as well as image guidance alone, are unlikely causes of the large discrepancy of toxicities. Whether the cause was inherent to the specific NTCP models used in this study needs to be verified by future investigations. Because IMRT is increasingly used clinically, it is important that appropriate NTCP model parameters are determined for this treatment modality.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Bifunctional Janus Particles as Multivalent Synthetic Nanoparticle Antibodies (SNAbs) for Selective Depletion of Target Cells

Monoclonal antibodies (mAb) have had a transformative impact on treating cancers and immune disorders. SNAbs However, their use is limited by high development time and monetary cost, manufacturing complexities, suboptimal pharmacokinetics, and availability of disease-specific targets. To address some of these challenges, we developed an entirely synthetic, multivalent, Janus nanotherapeutic platform, called Synthetic Nanoparticle Antibodies (SNAbs). SNAbs, with phage-display-identified cell-targeting ligands on one "face" and Fc-mimicking ligands on the opposite "face", were synthesized using a custom, multistep, solid-phase chemistry method. SNAbs efficiently targeted and depleted myeloid-derived immune-suppressor cells (MDSCs) from mouse-tumor and rat-trauma models, ex vivo. Systemic injection of MDSC-targeting SNAbs efficiently depleted circulating MDSCs in a mouse triple-negative breast cancer model, enabling enhanced T cell and Natural Killer cell infiltration into tumors. Finally, our results demonstrate that SNAbs are a versatile and effective functional alternative to mAbs, with advantages of a plug- and-play, cell-free manufacturing process, and high-throughput screening (HTS)-enabled library of potential targeting ligands.

60 APPLIED LIFE SCIENCES↗

SU-F-T-348: The Impact of Model Library Population On RapidPlan Based Dose-Volume Histograms (DVHs) Prediction for Rectal Cancer Patients Treated with Volumetric-Modulated Radiotherapy (VMAT)

Purpose: RapidPlan uses a library consisting of expert plans from different patients to create a model that can predict achievable dose-volume histograms (DVHs) for new patients. The goal of this study is to investigate the impacts of model library population (plan numbers) on the DVH prediction for rectal cancer patients treated with volumetric-modulated radiotherapy (VMAT) Methods: Ninety clinically accepted rectal cancer patients’ VMAT plans were selected to establish 3 models, named as Model30, Model60 and Model90, with 30,60, and 90 plans in the model training. All plans had sufficient target coverage and bladder and femora sparings. Additional 10 patients were enrolled to test the DVH prediction differences with these 3 models. The predicted DVHs from these 3 models were compared and analyzed. Results: Predicted V40 (Vx, percent of volume that received x Gy for the organs at risk) and Dmean (mean dose, cGy) of the bladder were 39.84±13.38 and 2029.4±141.6 for the Model30,37.52±16.00 and 2012.5±152.2 for the Model60, and 36.33±18.35 and 2066.5±174.3 for the Model90. Predicted V30 and Dmean of the left femur were 23.33±9.96 and 1443.3±114.5 for the Model30, 21.83±5.75 and 1436.6±61.9 for the Model60, and 20.31±4.6 and 1415.0±52.4 for the Model90.There were no significant differences among the 3 models for the bladder and left femur predictions. Predicted V40 and Dmean of the right femur were 19.86±10.00 and 1403.6±115.6 (Model30),18.97±6.19 and 1401.9±68.78 (Model60), and 21.08±7.82 and 1424.0±85.3 (Model90). Although a slight lower DVH prediction of the right femur was found on the Model60, the mean differences for V30 and mean dose were less than 2% and 1%, respectively. Conclusion: There were no significant differences among Model30, Model60 and Model90 for predicting DVHs on rectal patients treated with VMAT. The impact of plan numbers for model library might be limited for cancers with similar target shape.

60 APPLIED LIFE SCIENCES↗

Recursive Partitioning Analysis of Prognostic Factors for Survival in Patients With Advanced Cancer

Purpose: To construct a predictive model for survival of patients referred to the Rapid Response Radiotherapy Program using recursive partitioning (RP). Methods and Materials: We analyzed 16 factors characterizing patients with metastases at first referral to the Rapid Response Radiotherapy Program for palliative radiotherapy in 1999 for their effect on survival. They included age, primary cancer site, site of metastases, weight loss ({>=}10% during the past 6 months), Karnofsky performance status (KPS), interval from the first diagnosis of cancer to the first consultation at the Rapid Response Radiotherapy Program, analgesic consumption within the previous 24 h, and the nine symptom scores from the Edmonton Symptom Assessment Scale. We used RP to develop a predictive model of survival for patients referred in 1999, followed by temporal validation using patients referred in 2000, and external validation using patients referred in 2002 to another institution. Results: The model was able to separate patients into three groups with different durations of survival that were defined by (1) KPS >60; (2) KPS {<=}60 with bone metastases only; and (3) KPS {<=}60 with other metastases. The model performed moderately well when applied to the validation sets, but a major limitation was that it led to an unequal distribution of patients, with a small proportion of patients in the intermediate group. Conclusion: We have demonstrated that RP can be used to predict the survival of patients with advanced cancer. However, this model has no advantages compared with our published prognostic models that use the survival prediction scores and number of risk factors.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

A cross-study analysis of drug response prediction in cancer cell lines

Abstract To enable personalized cancer treatment, machine learning models have been developed to predict drug response as a function of tumor and drug features. However, most algorithm development efforts have relied on cross-validation within a single study to assess model accuracy. While an essential first step, cross-validation within a biological data set typically provides an overly optimistic estimate of the prediction performance on independent test sets. To provide a more rigorous assessment of model generalizability between different studies, we use machine learning to analyze five publicly available cell line-based data sets: National Cancer Institute 60, ancer Therapeutics Response Portal (CTRP), Genomics of Drug Sensitivity in Cancer, Cancer Cell Line Encyclopedia and Genentech Cell Line Screening Initiative (gCSI). Based on observed experimental variability across studies, we explore estimates of prediction upper bounds. We report performance results of a variety of machine learning models, with a multitasking deep neural network achieving the best cross-study generalizability. By multiple measures, models trained on CTRP yield the most accurate predictions on the remaining testing data, and gCSI is the most predictable among the cell line data sets included in this study. With these experiments and further simulations on partial data, two lessons emerge: (1) differences in viability assays can limit model generalizability across studies and (2) drug diversity, more than tumor diversity, is crucial for raising model generalizability in preclinical screening.

59 BASIC BIOLOGICAL SCIENCES↗

Hypofractionation: What Does It Mean for Prostate Cancer Treatment?

Purpose: Using current radiobiologic models and biologic parameters, we performed an exploratory study of the clinical consequences of hypofractionation in prostate cancer radiotherapy. Methods and Materials: Four hypofractionated treatment regimens were compared with standard fractionation of 2 Gy x 39 for prostate carcinoma using a representative set of anatomical structures. The linear-quadratic model and generalized equivalent uniform dose formalism were used to calculate normalized equivalent uniform dose (gEUD{sub 2}), from which tumor control probability and normal tissue complication probability were calculated, as well as 'complication-free tumor control probability' (P+). The robustness of the results was tested for various tumor alpha/beta values and broad interval of biologic parameters such as surviving fraction after a dose of 2 Gy (SF2). Results: A 2.5% and 5.8% decrease in NTCP for rectum and bladder, respectively, was predicted for the 6.5 Gy/fraction regimen compared with the 2 Gy/fraction. Conversely, TCP for hypofractionated regimens decreased significantly with increasing SF2 and alpha/beta. For tumor cells with SF2 = 0.4-0.5, P+ was superior for nearly all hypofractionated regimens even for alpha/beta values up to 6.5 Gy. For less responsive tumor cells (SF2 = 0.6), hypofractionation regimens were inferior to standard fractionation at much lower alpha/beta. Conclusion: For a sample set of anatomical structures, existing radiobiologic data and models predict improved clinical results from hypofractionation over standard fractionation not only for prostate carcinoma with low alpha/beta but also for high alpha/beta (up to 6.5 Gy) when SF2 < 0.5. Predicted results for specific patients may vary with individual anatomy, and large-scale clinical conclusions can be drawn only after performing similar analysis on an appropriate population of patients.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Confirmation of a Low {alpha}/{beta} Ratio for Prostate Cancer Treated by External Beam Radiation Therapy Alone Using a Post-Treatment Repeated-Measures Model for PSA Dynamics

Purpose: To estimate the {alpha}/{beta} ratio of prostate cancer treated with external beam radiation only by use of a model of long-term prostate-specific antigen (PSA) dynamics. Methods and Materials: Repeated measures of PSA from 5,093 patients from 6 institutions treated for localized prostate cancer by external beam radiation therapy (EBRT) without planned androgen deprivation were analyzed. A biphasic linear mixed model described the post-treatment evolution of PSA, rather than a conventional model of time to biochemical recurrence. The model was adjusted for standard prognostic factors (T stage, initial PSA level, and Gleason score) and cohort-specific effects. The radiation dose fractionation effect was estimated from the long-term rate of rise of PSA level. Results: Adjusted for other factors, total dose of EBRT and sum of squared doses per fraction were associated with long-term rate of change of PSA level (p = 0.0017 and p = 0.0003, respectively), an increase of each being associated with a lower rate of rise. The {alpha}/{beta} ratio was estimated at 1.55 Gy (95% confidence band, 0.46-4.52 Gy). This estimate was robust to adjustment of the linear mixed model. Conclusions: By analysis of a large EBRT-only cohort along with a method that uses all the repeated measures of PSA after the end of treatment, a low and precise {alpha}/{beta} was estimated. These data support the use of hypofractionation at fractional doses up to 2.8 Gy but cannot presently be assumed to accurately represent higher doses per fraction.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Lung Cancer in the Mayak Workers Cohort: Risk Estimation and Uncertainty Analysis

The workers at the Mayak nuclear facility near Ozyorsk, Russia are a primary source of information about exposure to radiation at low-dose rates, since they were subject to protracted exposures to external gamma rays and to internal exposures from plutonium inhalation. Here we re-examine lung cancer mortality rates and assess the effects of external gamma and internal plutonium exposures using recently developed Monte Carlo dosimetry systems. Using individual lagged mean annual lung doses computed from the dose realizations, we fit excess relative risk (ERR) models to the lung cancer mortality data for the Mayak Workers Cohort using risk-modeling software. We then used the corrected information matrix (CIM) approach to widen the confidence intervals of ERR by taking into account the uncertainty in doses represented by multiple realizations from the Monte Carlo dosimetry systems. Findings of this work revealed that there were 930 lung cancer deaths during follow-up. Plutonium lung doses (but not gamma doses) were generally higher in the new dosimetry systems than those used in the previous analysis. This led to a reduction in the risk per unit dose compared to prior estimates. The estimated ERR/Gy for external gamma-ray exposure was 0.19 (95% CI: 0.07 to 0.31) for both sexes combined, while the ERR/Gy for internal exposures based on mean plutonium doses were 3.5 (95% CI: 2.3 to 4.6) and 8.9 (95% CI: 3.4 to 14) for males and females at attained age 60. Accounting for uncertainty in dose had little effect on the confidence intervals for the ERR associated with gamma-ray exposure, but had a marked impact on confidence intervals, particularly the upper bounds, for the effect of plutonium exposure [adjusted 95% CIs: 1.5 to 8.9 for males and 2.7 to 28 for females]. In conclusion, lung cancer rates increased significantly with both external gamma-ray and internal plutonium exposures. Accounting for the effects of dose uncertainty markedly increased the width of the confidence intervals for the plutonium dose response but had little impact on the external gamma dose effect estimate. Adjusting risk estimate confidence intervals using CIM provides a solution to the important problem of dose uncertainty. This work demonstrates, for the first time, that it is possible and practical to use our recently developed CIM method to make such adjustments in a large cohort study.

uncertainty, radiation risk estimation, Mayak Prod↗

TU-F-BRF-08: Intensity Modulation Expanded to Spatio-Temporal Space; IMRT Combined with Optimal Fractionation Schedule

Purpose: Past efforts to improve the therapeutic ratio have focused on a spatial approach where highly conformal radiation dose is given to tumors while minimizing dose to normal tissues, e.g., IMRT, VMAT, and IGRT. However, the fractionation schedule, i.e., a temporal approach to radiotherapy, has been largely overlooked so far in maximizing the therapeutic ratio. We establish a rigorous mathematical spatio-temporal approach to systematically investigate the feasibility and potential benefits of simultaneously optimizing radiation dose distribution in space and time. Methods: Stochastic control formalism is constructed to maximize the average tumor BED by choosing an optimal radiation dose distribution for an optimal number of fractions subject to normal tissue BED constraints. Three separate simulations are run on two groups of phantom cases; 5 cases with prostate cancer and 5 cases with head-and-neck cancer. (1) Conventional IMRT with 70Gy/35fx for head-and-neck, and 81Gy/45fx for prostate, (2) IMRT is done independently from the fractionation schedule optimization (S-model), (3) integrated spatio-temporal approach (I-model) where radiation intensities are simultaneously optimized for the first time ever in space and time. Final tumor BEDs from the three trials are compared in prostate and head-and-neck cases. Results: Numerical simulations show that final tumor BED from I-model is 20–90% larger than conventional IMRT, and 20-50% larger than S-model for head-and-neck cancer with α/β=10 and Tdouble=2–50 days. Final tumor BED from I-model is also 90–140% larger than conventional IMRT, and 20–30% larger than S-model for prostate cancer with α/β=2 and Tdouble=5–80 days. Conclusion: Our spatio-temporal optimization of radiotherapy allows an expansion of search space for the optimal treatment plans to include the temporal distribution of radiation dose in addition to the spatial distribution. Such spatio-temporal approach shows great potential to improve the therapeutic ratio, particularly with fast growing, aggressive tumors with short doubling time and/or low α/β.

60 APPLIED LIFE SCIENCES↗

Model-based Hyperparameter Optimization of Convolutional Neural Networks for Information Extraction from Cancer Pathology Reports on HPC

Finding optimal hyperparameters is necessary to identify the best performing deep learning models but the process is costly. In this paper, we applied model-based optimization, also known as Bayesian optimization, using the CANDLE framework implemented on a High-Performance Computing environment. As a use case we selected information extraction from cancer pathology reports using a multi-task convolutional neural network, and hierarchical convolutional attention network to be optimized. We utilized a synthesized text corpus of 8,000 training cases and 2,000 validation cases with four types of clinical task labels including primary cancer site, laterality, behavior, and histological grade. We demonstrated that hyperparameter optimization using the CANDLE framework is a feasible approach with respect to both scalability and clinical task performance.

Yoon, Hong-Jun↗