Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “false negative”

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 145 records · Page 8

Cross-Combined Composite Sampling Designs for Identification of Elevated Regions

Analyzing soils for contaminants can be costly. Generally, discrete samples are gathered from within a study area, analyzed by a laboratory and the results are used in a site-specific statistical analysis. Because of the heterogeneities that exist in soil samples within study areas, a large amount of variability and skewness may be present in the sample population. This necessitates collecting a large number of samples to obtain reliable inference on the mean contaminant concentration and to understand the spatial patterns for future remediation. Composite or Incremental sampling is a commonly applied method for gathering multiple discrete samples and physically combining them, such that each combination of discrete samples requires a single laboratory analysis, which reduces cost and can improve the estimates of the mean concentration. While incremental sampling can reduce cost and improve mean estimates, current implementations do not readily facilitate the characterization of spatial patterns or the detection of elevated regions within study areas. The methods we present in this work provide efficient estimation and inference for the mean contaminant concentration over the entire spatial area and enable the identification of high contaminant regions within the area of interest. We develop sample design methodologies that explicitly define the characteristics of these designs (such as sample grid layout) and quantify the number of incremental samples that must be obtained under a design criteria to control false positive and false negative (Type I and II) decision errors. We present the sample design theory and specifications as well as results on simulated and real data.

Hathaway, John E↗

Crowdsourced reliable labeling of safety-rule violations on images of complex construction scenes for advanced vision-based workplace safety

Construction workplace hazard detection requires engineers to analyze scenes manually against many safety rules, which is time-consuming, labor-intensive, and error-prone. Computer vision algorithms are yet to achieve reliable discrimination of anomalous and benign object relations underpinning safety violation detections. Recently developed deep learning-based computer vision algorithms need tens of thousands of images, including labels of the safety rules violated, in order to train deep-learning networks for acquiring spatiotemporal reasoning capacity in complex workplaces. Such training processes need human experts to label images and indicate whether the relationship between the worker, resource, and equipment in the scenes violate spatiotemporal arrangement rules for safe and productive operations. False alarms in those manual labels (labeling no-violation images as having violations) can significantly mislead the machine learning process and result in computer vision models that produce inaccurate hazard detections. Compared with false alarms, another type of mislabels, false negatives (labeling images having violations as “no violations”), seem to have fewer impacts on the reliability of the trained computer vision models.

97 MATHEMATICS AND COMPUTING↗

Computing the binding affinity of a ligand buried deep inside a protein with the hybrid steered molecular dynamics

Computing the ligand-protein binding affinity (or the Gibbs free energy) with chemical accuracy has long been a challenge for which many methods/approaches have been developed and refined with various successful applications. False positives and, even more harmful, false negatives have been and still are a common occurrence in practical applications. Inevitable in all approaches are the errors in the force field parameters we obtain from quantum mechanical computation and/or empirical fittings for the intra- and inter-molecular interactions. These errors propagate to the final results of the computed binding affinities even if we were able to perfectly implement the statistical mechanics of all the processes relevant to a given problem. And they are actually amplified to various degrees even in the mature, sophisticated computational approaches. In particular, the free energy perturbation (alchemical) approaches amplify the errors in the force field parameters because they rely on extracting the small differences between similarly large numbers. In this paper, we develop a hybrid steered molecular dynamics (hSMD) approach to the difficult binding problems of a ligand buried deep inside a protein. Sampling the transition along a physical (not alchemical) dissociation path of opening up the binding cavity- -pulling out the ligand- -closing back the cavity, we can avoid the problem of error amplifications by not relying on small differences between similar numbers. We tested this new form of hSMD on retinol inside cellular retinol-binding protein 1 and three cases of a ligand (a benzylacetate, a 2-nitrothiophene, and a benzene) inside a T4 lysozyme L99A/M102Q(H) double mutant. In all cases, we obtained binding free energies in close agreement with the experimentally measured values. This indicates that the force field parameters we employed are accurate and that hSMD (a brute force, unsophisticated approach) is free from the problem of error amplification suffered by many sophisticated approaches in the literature.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Development and testing of a performance evaluation methodology to assess the reliability of occupancy sensor systems in residential buildings

With the emergence of advanced occupancy sensor technologies to better detect occupancy in buildings, a universal methodology and metrics are required to evaluate and report sensor systems’ reliability and compare the performance across multiple sensor systems. Herein this research presents a methodology to assess the reliability of occupancy sensor systems in residential buildings in a controlled laboratory environment, including both “typical” and “failure” testing scenarios. The developed methodology was then implemented to evaluate a novel occupancy detection sensor system’s reliability. “Typical” testing evaluates the overall accuracy of the sensor system, which suggest how reliable the occupancy sensor system is over time. Results show that on average, the precision and recall are 0.75 and 0.70, indicating similar numbers of false positives and false negatives across the dataset. The overall accuracy of the tested sensor system was 62.4% to 76.4%. Failure testing results indicate whether there are influential variables impacting the sensor performance. For the tested sensor system, the number of occupants, presence of large objects, presence of interior light sources, and number of doors are not influential, while lighting level, location of occupants, additional door in the entry/exit area, and having the TV on are variables determined to impact the sensor system performance.

47 OTHER INSTRUMENTATION↗

Long-Term field testing of the accuracy and HVAC energy savings potential of occupancy presence sensors in A Single-Family home

The energy-saving potential of occupancy-centric smart thermostats has been extensively explored in simulations but lacked field testing for energy savings quantification and sensor performance assessment in real buildings. This paper presents a long-term field study conducted in a single-family home in Texas, U.S. to evaluate the performance of occupancy-centric controls (OCC) of HVAC (heating, ventilation, and air-conditioning) system in terms of energy savings, sensor accuracy, and impact on electric peak demand. The test site was equipped with a commercial off-the-shelf (COTS) smart thermostat and multiple occupancy presence sensors for OCC implementation. Additionally, a sub-metering system was installed to monitor electricity consumption of various end-use equipment, including the HVAC system. A supplementary device was installed to track the ground-truth occupancy for the accuracy evaluation of the occupancy presence sensor. Scenarios of baseline and OCC controls were alternated weekly over the 20-month testing period. The results indicated an effective OCC execution, as evidenced by indoor temperature profiles. During the 2023 cooling season, OCC achieved total energy savings of 1,958 kWh, corresponding to a 17.6% energy savings ratio. Under certain conditions, daily HVAC energy savings reached as high as 17 kWh, with a savings ratio of 35%. Sensor performance showed an overall accuracy of 83.8%, a False Positive Rate (FPR) of 12.8%, and a False Negative Rate (FNR) of 47.4%. A key limitation was the sensor’s inability to detect stationary occupants during sleep, leading to a midnight FNR of nearly 100% and significantly compromising thermal comfort. Additionally, the implementation of OCC resulted in extended periods of high electricity demand on summer afternoons, affecting occupant’s thermal comfort and posing potential challenges to community-level grid operations if OCC were widely adopted. Furthermore, this study addresses a critical research gap by empirically investigating energy-saving potential and occupancy sensor performance in residential buildings. Through a comprehensive field-testing study, the research examines the interrelationship between sensor accuracy, energy savings, and thermal comfort, an area that has received limited attention in the current literature.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Examination of lead concentrations in new decorative enamel paints in four countries with different histories of activity in lead paint regulation

Paints with high lead concentrations (ppm) continue to be sold around the world in many developing countries and those with economies in transition representing a major preventable environmental health hazard that is being increased as the economies expand and paint consumption is increasing. Prior lead paint testing had been performed in Brazil and India and these countries were selected to examine the impact of a new regulatory limit in Brazil and the impact of efforts of non-governmental organizations and others to stop the use of lead compounds in manufacturing paints. Armenia and Kazakhstan, in Central Asia, were selected because no information on lead concentration in those regions was available, no regulatory activities were evident and non-governmental organizations in the IPEN network were available to participate. Another objective of this research was to evaluate the lead loading (µg/cm{sup 2}) limit determined by X-Ray Fluorescence (XRF) for areas on toys that are too small to obtain a sample of sufficient size for laboratory analysis. The lead concentrations in more than three-fourths of the paints from Armenia and Kazakhstan exceeded 90 ppm, the limit in the United States, and 600 ppm, the limit in Brazil. The percentages were about one-half as high in Brazil and India. The average concentration in paints purchased in Armenia, 25,000 ppm, is among the highest that has been previously reported, that in Kazakhstan, 15,700 ppm, and India, 16,600, about median. The average concentration in Brazil, 5600 ppm, is among the lowest observed. Paints in Brazil that contained an average of 36,000 ppm before the regulatory limit became effective were below detection (<9 ppm) in samples collected in the current study. The lack of any apparent public monitoring of paint lead content as part of regulatory enforcement makes it difficult to determine whether the regulation was a major factor contributing to the decline in lead use in these paints. Using data from the current study and those available from other studies 24 of 28 paints from major brands in India decreased from high concentrations to 90 ppm or lower. Since lead concentrations in golden yellow paints from these brands were found to decrease to ≤90 ppm, it is possible that all 28 of these paints now contain ≤90 ppm since yellow paints usually have the highest lead concentrations. Other brands in Brazil and India that have been analyzed only one time had lead concentrations up to 59,000 ppm and 134,000 ppm, respectively. Less than one-third of the paints had notations on their labels with information about lead content and these were sometimes inaccurate. The label from one brand indicating “no added lead” contained paint with 134,000 ppm lead, the highest found in this study. Three percent (3 of 98) of the paints with surface lead loading that did not exceed 2 µg/cm{sup 2}, the limit established by the Consumer Product Safety Improvement Act for small areas on toys, contained greater than 90 ppm lead and thus were false negatives. Of the new paint samples that contained ≤600 ppm, 88% contained ≤90 ppm. Of the samples that contained ≤90 ppm, 97% contained ≤45 ppm and 92% contained ≤15 ppm. Based on these data it appears to be technically feasible to manufacture paints containing ≤90 ppm and in many cases to produce paints that have lead concentrations that do not exceed 15 ppm. - Highlights: • Brazilian paints with high lead (avg. 36,000 ppm) contained <9 ppm after regulation. • Lead in major Indian paint brands decreased to low levels by efforts of NGOs/others. • Most paints with lead ≤600 ppm also are ≤90 ppm and many also ≤15 ppm. • At the CPSIA limit of 2 µg Pb/cm{sup 2} for small areas on toys, 3% exceeded 90 ppm lead. • Most paint brands in Armenia and Kazakhstan based/manufactured outside the country.

54 ENVIRONMENTAL SCIENCES↗

Evaluation and comparison of benchmark QSAR models to predict a relevant REACH endpoint: The bioconcentration factor (BCF)

The bioconcentration factor (BCF) is an important bioaccumulation hazard assessment metric in many regulatory contexts. Its assessment is required by the REACH regulation (Registration, Evaluation, Authorization and Restriction of Chemicals) and by CLP (Classification, Labeling and Packaging). We challenged nine well-known and widely used BCF QSAR models against 851 compounds stored in an ad-hoc created database. The goodness of the regression analysis was assessed by considering the determination coefficient (R{sup 2}) and the Root Mean Square Error (RMSE); Cooper's statistics and Matthew's Correlation Coefficient (MCC) were calculated for all the thresholds relevant for regulatory purposes (i.e. 100 L/kg for Chemical Safety Assessment; 500 L/kg for Classification and Labeling; 2000 and 5000 L/kg for Persistent, Bioaccumulative and Toxic (PBT) and very Persistent, very Bioaccumulative (vPvB) assessment) to assess the classification, with particular attention to the models' ability to control the occurrence of false negatives. As a first step, statistical analysis was performed for the predictions of the entire dataset; R{sup 2}>0.70 was obtained using CORAL, T.E.S.T. and EPISuite Arnot–Gobas models. As classifiers, ACD and log P-based equations were the best in terms of sensitivity, ranging from 0.75 to 0.94. External compound predictions were carried out for the models that had their own training sets. CORAL model returned the best performance (R{sup 2}{sub ext}=0.59), followed by the EPISuite Meylan model (R{sup 2}{sub ext}=0.58). The latter gave also the highest sensitivity on external compounds with values from 0.55 to 0.85, depending on the thresholds. Statistics were also compiled for compounds falling into the models Applicability Domain (AD), giving better performances. In this respect, VEGA CAESAR was the best model in terms of regression (R{sup 2}=0.94) and classification (average sensitivity>0.80). This model also showed the best regression (R{sup 2}=0.85) and sensitivity (average>0.70) for new compounds in the AD but not present in the training set. However, no single optimal model exists and, thus, it would be wise a case-by-case assessment. Yet, integrating the wealth of information from multiple models remains the winner approach. - Highlights: • REACH encourages the use of in silico methods in the assessment of chemicals safety. • The performances of nine BCF models were evaluated on a benchmark database of 851 chemicals. • We compared the models on the basis of both regression and classification performance. • Statistics on chemicals out of the training set and/or within the applicability domain were compiled. • The results show that QSAR models are useful as weight-of-evidence in support to other methods.

54 ENVIRONMENTAL SCIENCES↗

Planning, design and logistics of a decision analysis study: The FBI/Ames study involving forensic firearms examiners

This paper describes design and logistical aspects of a decision analysis study to assess the performance of qualified firearms examiners working in accredited laboratories in the United States in terms of accuracy (error rate), repeatability, and reproducibility of decisions involving comparisons of fired bullets and cartridge cases. The purpose of the study was to validate current practice of the forensic discipline of firearms/toolmarks (F/T) examination. It elicited error rate data by counting the number of false positive and false negative conclusions. Preceded by the experimental design, decisions, and logistics described herein, testing was ultimately administered 173 qualified, practicing F/T examiners in public and private crime laboratories. The first round of testing evaluated accuracy, while two subsequent rounds evaluated repeatability and reproducibility of examiner conclusions. This project expands on previous studies by involving many F/T examiners in challenging comparisons and by executing the study in the recommended double-blind format.

99 GENERAL AND MISCELLANEOUS↗

Impact of {sup 18}F-Fluorodeoxyglucose Positron Emission Tomography Before and After Definitive Radiation Therapy in Patients With Apparently Solitary Plasmacytoma

Purpose: To evaluate the impact of {sup 18}F-fluorodeoxyglucose positron emission tomography (FDG-PET) on management of patients with apparently isolated plasmacytoma. Methods and Materials: Twenty-one patients with apparently solitary plasmacytoma who underwent FDG-PET for staging or restaging were identified from a central PET database. They were either candidates for or had received definitive radiation therapy (RT). Results: Seventeen patients had initial staging scans for bone (n = 11) or soft tissue (n = 6) plasmacytomas, and 11 had PET scans after RT. Only 1 of 14 known untreated sites of plasmacytoma was not identified on staging PET (lesion sensitivity = 93%). Three plasmacytomas were excised before PET. Staging PET influenced management in 6 of 17 patients (35%) by showing multiple myeloma (n = 1), discouraging RT after complete resection (n = 1), excluding plasmacytoma at a second site (n = 1), by increasing RT fields (n = 2), or by suggesting sarcoidosis (n = 1). Fifteen of 17 patients with initial staging PET scans received definitive RT. Restaging PET scans after RT showed complete metabolic response in 8 of 11 cases and progressive disease in 2. Two patients with either no response or partial metabolic response had late responses. Staging sestamibi and PET scans were concordant in five of six occasions (one sestamibi scan was false negative). Conclusions: FDG-PET has value for staging and RT planning in plasmacytoma and potentially could have a role in response-assessment after RT. Slow resolution of FDG uptake posttreatment does not necessarily imply an adverse prognosis.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Indications for Pelvic Nodal Treatment in Prostate Cancer Should Change. Validation of the Roach Formula in a Large Extended Nodal Dissection Series

Purpose: Previous studies have criticized the predicting ability of the Roach formula in assessing the risk of lymph node invasion (LNI) in contemporary patients with prostate cancer (PCa) due to a significant overestimation of LNI rates. However, all those studies included patients treated with limited pelvic lymph node dissection (PLND), which is associated with high rates of false negative findings. We hypothesized that the Roach formula is still an accurate tool for LNI predictions if an extended PLND (ePLND) is performed. Methods and Materials: We included 3,115 consecutive patients treated with radical prostatectomy and ePLND between 2000 and 2010 at a single tertiary referral center. Extended PLND consisted of removal of obturator, external iliac, and hypogastric lymph nodes. We externally validated the Roach formula by using the area under the receiver operating characteristics curve and calibration plot method. Moreover, we tested the performance characteristics of different formula-generated cutoff values ranging from 1% to 20%. Results: The accuracy of the Roach formula was 80.3%. The calibration showed only a minor underestimation of the LNI risk in high-risk patients (6.7%). According to the Roach formula, the use of 15% cut off would have allowed 74.2% (2,311/3,115) of patients to avoid nodal irradiation, while up to 32.7% (111/336) of all patients with LNI would have been missed. When the cut off was lowered to 6%, nodal treatment would have been spared in 1,541 (49.5%) patients while missing 41 LNI patients. The sensitivity, specificity, and negative predictive values associated with the 6% cut off were 87.9%, 54%, and 97.3%, respectively. Conclusions: The Roach formula is still accurate and does not overestimate the rate of LNI in contemporary prostate cancer patients if they are treated with ePLND. However, the recommended cut off of 15% would miss approximately one-third of patients with LNI. Based on our results, the cut off should be lowered to 6%.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Evaluation of 4-dimensional Computed Tomography to 4-dimensional Cone-Beam Computed Tomography Deformable Image Registration for Lung Cancer Adaptive Radiation Therapy

Purpose: To evaluate 2 deformable image registration (DIR) algorithms for the purpose of contour mapping to support image-guided adaptive radiation therapy with 4-dimensional cone-beam CT (4DCBCT). Methods and Materials: One planning 4D fan-beam CT (4DFBCT) and 7 weekly 4DCBCT scans were acquired for 10 locally advanced non-small cell lung cancer patients. The gross tumor volume was delineated by a physician in all 4D images. End-of-inspiration phase planning 4DFBCT was registered to the corresponding phase in weekly 4DCBCT images for day-to-day registrations. For phase-to-phase registration, the end-of-inspiration phase from each 4D image was registered to the end-of-expiration phase. Two DIR algorithms—small deformation inverse consistent linear elastic (SICLE) and Insight Toolkit diffeomorphic demons (DEMONS)—were evaluated. Physician-delineated contours were compared with the warped contours by using the Dice similarity coefficient (DSC), average symmetric distance, and false-positive and false-negative indices. The DIR results are compared with rigid registration of tumor. Results: For day-to-day registrations, the mean DSC was 0.75 ± 0.09 with SICLE, 0.70 ± 0.12 with DEMONS, 0.66 ± 0.12 with rigid-tumor registration, and 0.60 ± 0.14 with rigid-bone registration. Results were comparable to intraobserver variability calculated from phase-to-phase registrations as well as measured interobserver variation for 1 patient. SICLE and DEMONS, when compared with rigid-bone (4.1 mm) and rigid-tumor (3.6 mm) registration, respectively reduced the average symmetric distance to 2.6 and 3.3 mm. On average, SICLE and DEMONS increased the DSC to 0.80 and 0.79, respectively, compared with rigid-tumor (0.78) registrations for 4DCBCT phase-to-phase registrations. Conclusions: Deformable image registration achieved comparable accuracy to reported interobserver delineation variability and higher accuracy than rigid-tumor registration. Deformable image registration performance varied with the algorithm and the patient.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Extended Field Intensity Modulated Radiation Therapy With Concomitant Boost for Lymph Node–Positive Cervical Cancer: Analysis of Regional Control and Recurrence Patterns in the Positron Emission Tomography/Computed Tomography Era

Purpose: Positron emission tomography/computed tomography (PET/CT) is commonly used for nodal staging in locally advanced cervical cancer; however the false negative rate for para-aortic disease are 20% to 25% in PET-positive pelvic nodal disease. Unless surgically staged, pelvis-only treatment may undertreat para-aortic disease. We have treated patients with PET-positive nodes with extended field intensity modulated radiation therapy (IMRT) to address the para-aortic region prophylactically with concomitant boost to involved nodes. The purpose of this study was to assess regional control rates and recurrence patterns. Methods and Materials: Sixty-one patients with cervical cancer (stage IBI-IVA) diagnosed from 2003 to 2012 with PET-avid pelvic nodes treated with extended field IMRT (45 Gy in 25 fractions with concomitant boost to involved nodes to a median of 55 Gy in 25 fractions) with concurrent cisplatin and brachytherapy were retrospectively analyzed. The nodal location was pelvis-only in 41 patients (67%) and pelvis + para-aortic in 20 patients (33%). There were a total of 179 nodes, with a median number of positive nodes of 2 (range, 1-16 nodes) per patient and a median nodal size of 1.8 cm (range, 0.7-4.5 cm). Response was assessed by PET/CT at 12 to 16 weeks. Results: Complete clinical and imaging response at the first follow-up visit was seen in 77% of patients. At a mean follow-up time of 29 months (range, 3-116 months), 8 patients experienced recurrence. The sites of persistent/recurrent disease were as follows: cervix 10 (16.3%), regional nodes 3 (4.9%), and distant 14 (23%). The rate of para-aortic failure in patients with pelvic-only nodes was 2.5%. There were no significant differences in recurrence patterns by the number/location of nodes, largest node size, or maximum node standardized uptake value. The rate of late grade 3+ adverse events was 4%. Conclusions: Extended field IMRT was well tolerated and resulted in low regional recurrence in node-positive cervical cancer. The dose of 55 Gy in 25 fractions was effective in eradicating disease in involved nodes, with acceptable late adverse events. Distant metastasis is the predominant mode of failure, and the OUTBACK trial may challenge the presented paradigms.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Patterns of Local-Regional Management Following Neoadjuvant Chemotherapy in Breast Cancer: Results From ACOSOG Z1071 (Alliance)

Purpose: American College of Surgeons Oncology Group Z1071 was a prospective trial evaluating the false negative rate of sentinel lymph node (SLN) surgery after neoadjuvant chemotherapy (NAC) in breast cancer patients with initial node-positive disease. Radiation therapy (RT) decisions were made at the discretion of treating physicians, providing an opportunity to evaluate variability in practice patterns following NAC. Methods and Materials: Of 756 patients enrolled from July 2009 to June 2011, 685 met all eligibility requirements. Surgical approach, RT, and radiation field design were analyzed based on presenting clinical and pathologic factors. Results: Of 401 node-positive patients, mastectomy was performed in 148 (36.9%), mastectomy with immediate reconstruction in 107 (26.7%), and breast-conserving surgery (BCS) in 146 patients (36.4%). Of the 284 pathologically node-negative patients, mastectomy was performed in 84 (29.6%), mastectomy with immediate reconstruction in 69 (24.3%), and BCS in 131 patients (46.1%). Bilateral mastectomy rates were higher in women undergoing reconstruction than in those without (66.5% vs 32.2%, respectively, P<.0001). Use of internal mammary RT was low (7.8%-11.2%) and did not differ between surgical approaches. Supraclavicular RT rate ranged from 46.6% to 52.2% and did not differ between surgical approaches but was omitted in 193 or 408 node-positive patients (47.3%). Rate of axillary RT was more frequent in patients who remained node-positive (P=.002). However, 22% of patients who converted to node-negative still received axillary RT. Post-mastectomy RT was more frequently omitted after reconstruction than mastectomy (23.9% vs 12.1%, respectively, P=.002) and was omitted in 19 of 107 patients (17.8%) with residual node-positive disease in the reconstruction group. Conclusions: Most clinically node-positive patients treated with NAC undergoing mastectomy receive RT. RT is less common in patients undergoing reconstruction. There is wide variability in RT fields. These practice patterns conflict with expert recommendations and ongoing trial guidelines. There is a significant need for greater uniformity and guidelines regarding RT following NAC.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Detection of Local Cancer Recurrence After Stereotactic Ablative Radiation Therapy for Lung Cancer: Physician Performance Versus Radiomic Assessment

Purpose: Stereotactic ablative radiation therapy (SABR) is a guideline-specified treatment option for early-stage lung cancer. However, significant posttreatment fibrosis can occur and obfuscate the detection of local recurrence. The goal of this study was to assess physician ability to detect timely local recurrence and to compare physician performance with a radiomics tool. Methods and Materials: Posttreatment computed tomography (CT) scans (n=182) from 45 patients treated with SABR (15 with local recurrence matched to 30 with no local recurrence) were used to measure physician and radiomic performance in assessing response. Scans were individually scored by 3 thoracic radiation oncologists and 3 thoracic radiologists, all of whom were blinded to clinical outcomes. Radiomic features were extracted from the same images. Performances of the physician assessors and the radiomics signature were compared. Results: When taking into account all CT scans during the whole follow-up period, median sensitivity for physician assessment of local recurrence was 83% (range, 67%-100%), and specificity was 75% (range, 67%-87%), with only moderate interobserver agreement (κ = 0.54) and a median time to detection of recurrence of 15.5 months. When determining the early prediction of recurrence within <6 months after SABR, physicians assessed the majority of images as benign injury/no recurrence, with a mean error of 35%, false positive rate (FPR) of 1%, and false negative rate (FNR) of 99%. At the same time point, a radiomic signature consisting of 5 image-appearance features demonstrated excellent discrimination, with an area under the receiver operating characteristic curve of 0.85, classification error of 24%, FPR of 24%, and FNR of 23%. Conclusions: These results suggest that radiomics can detect early changes associated with local recurrence that are not typically considered by physicians. This decision support system could potentially allow for early salvage therapy of patients with local recurrence after SABR.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Evaluation of Preoperative and Intraoperative Mobile Gamma Camera Imaging in Sentinel Lymph Node Biopsy for Melanoma Independent of Preoperative Lymphoscintigraphy

Sentinel lymph node biopsy (SLNB) is standard practice for staging cutaneous melanoma. High false negative rates have increased interest in adjunctive techniques for localizing SLNs. Mobile gamma cameras (MGCs) represent potential tools to enhance SLNB performance. Twenty eligible melanoma patients underwent 99mTc sulfur colloid injection and standard lymphoscintigraphy with a fixed gamma camera (FGC). A survey using a 20 cm square MGC, performed immediately preoperatively by the study surgeon, was used to establish an operative plan while blinded to the FGC results. Subsequently, SLNB was performed using a gamma probe and a novel 6 cm diameter handheld MGC. A total of 24 SLN basins were detected by FGC. Prior to unblinding, all 24 basins were identified with the preoperative MGC, and the operative plan established by preoperative MGC imaging was confirmed accurate by review of the FGC images. All individual SLNs were identified during intraoperative MGC imaging, and in 5/24 (21%) cases, additional clinical information was obtained from the MGC. Preoperative MGC images provide information consistent with FGC images for planning SLNB and in some cases provides additional information that aided in surgical decision-making.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Radial feature descriptors for cell classification and recommendation

This paper introduces computational tools for cell classification into normal and abnormal, as well as content-based-image-retrieval (CBIR) for cell recommendation. It also proposes the radial feature descriptors (RFD), which define evenly interspaced segments around the nucleus, and proportional to the convexity of the nuclear boundary. Experiments consider Herlev and CRIC image databases as input to classification via Random Forest and bootstrap; we compare 14 different feature sets by means of False Negative Rate (FNR) and Kappa (κ), obtaining FNR =0.02 and κ=0.89 for Herlev, and FNR =0.14 and κ=0.78 for CRIC. Next, we sort and rank cell images using convolutional neural networks and evaluate performance with the Mean Average Precision (MAP), achieving MAP =0.84 and MAP =0.82 for Herlev and CRIC, respectively. Finally, cell classification show encouraging results regarding RFD, including its sensitivity to intensity variation around the nuclear membrane as it bypasses cytoplasm segmentation.

47 OTHER INSTRUMENTATION↗

Initial experience of ArcCHECK and 3DVH software for RapidArc treatment plan verification

The purpose of this study was to perform delivery quality assurance with ArcCHECK and 3DVH system (Sun Nuclear, FL) and to evaluate the suitability of this system for volumetric-modulated arc therapy (VMAT) (RapidArc [RA]) verification. This software calculates the delivered dose distributions in patients by perturbing the calculated dose using errors detected in fluence or planar dose measurements. The device is tested to correlate the gamma passing rate (%GP) and the composite dose predicted by 3DVH software. A total of 28 patients with prostate cancer who were treated with RA were analyzed. RA treatments were delivered to a diode array phantom (ArcCHECK), which was used to create a planned dose perturbation (PDP) file. The 3DVH analysis used the dose differences derived from comparing the measured dose with the treatment planning system (TPS)-calculated doses to perturb the initial TPS-calculated dose. The 3DVH then overlays the resultant dose on the patient's structures using the resultant “PDP” beams. Measured dose distributions were compared with the calculated ones using the gamma index (GI) method by applying the global (Van Dyk) normalization and acceptance criteria, i.e., 3%/3 mm. Paired differences tests were used to estimate statistical significance of the differences between the composite dose calculated using 3DVH and %GP. Also, statistical correlation by means of logistic regression analysis has been analyzed. Dose-volume histogram (DVH) analysis for patient plans revealed small differences between treatment plan calculations and 3DVH results for organ at risk (OAR), whereas planning target volume (PTV) of the measured plan was systematically higher than that predicted by the TPS. The t-test results between the planned and the estimated DVH values showed that mean values were incomparable (p < 0.05). The quality assurance (QA) gamma analysis 3%/3 mm showed that in all cases there were only weak-to-moderate correlations (Pearson r: 0.12 to 0.74). Moreover, clinically relevant differences increased with increasing QA passing rate, indicating that some of the largest dose differences occurred in the cases of high QA passing rates, which may be called “false negatives.” The clinical importance of any disagreement between the measured and the calculated dose is often difficult to interpret; however, beam errors (either in delivery or in TPS calculation) can affect the effectiveness of the patient dose. Further research is needed to determinate the role of a PDP-type algorithm to accurately estimate patient dose effect.

61 RADIATION PROTECTION AND DOSIMETRY↗

Jensen–Shannon divergence based novel loss functions for Bayesian neural networks

Bayesian neural networks (BNNs) are state-of-the-art machine learning methods that can naturally regularize and systematically quantify uncertainties using their stochastic parameters. Kullback–Leibler (KL) divergence-based variational inference used in BNNs suffer from unstable optimization and challenges in approximating light-tailed posteriors due to the unbounded nature of the KL divergence. To resolve these issues, we formulate a novel loss function for BNNs based on a new modification to the generalized Jensen–Shannon (JS) divergence, which is bounded. In addition, we propose a Geometric JS divergence-based loss, which is computationally efficient since it can be evaluated analytically. We found that the JS divergence-based variational inference is intractable, and hence employed a constrained optimization framework to formulate these losses. Our theoretical analysis and empirical experiments on multiple regression and classification data sets suggest that the proposed losses perform better than the KL divergence-based loss, especially when the data sets are noisy or biased. Specifically, there are approximately 5% and 8% improvements in accuracy for a noise-added CIFAR-10 dataset and a regression dataset, respectively. There is about 13% reduction in false negative predictions of a biased histopathology dataset. Additionally, we quantify and compare the uncertainty metrics for the regression and classification tasks.

97 MATHEMATICS AND COMPUTING↗