Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “ASSISTED”

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 991 records · Page 55

Scaled-Up Fabrication of Thin-Walled ZK60 Tubing using Shear Assisted Processing and Extrusion (ShAPE)

Shear Assisted Processing and Extrusion (ShAPE) has been scaled-up and applied to direct extrusion of thin-walled magnesium tubing. Using ShAPE, billets of ZK60A-T5 were directly extruded into round tubes having an outer diameter of 50.8 mm and wall thickness of 1.52 mm. The severe shearing conditions inherent to ShAPE resulted in microstructural refinement with an average grain size of 3.8μm measured at the midpoint of the tube wall. Tensile testing per ATSM E-8 on specimens oriented parallel to the extrusion direction gave an ultimate tensile strength of 254.4 MPa and elongation of 20.1%. Specimens tested perpendicular to the extrusion direction had an ultimate tensile strength of 297.2 MPa and elongation of 25.0%. Due to material flow effects resulting from the simultaneous linear and rotational shear intrinsic to ShAPE, ram force and electrical power consumption during extrusion were just 40 kN and 11.5 kW respectively. This represents a significant reduction in ram force and power consumption compared to conventional extrusion. As such, there is potential for ShAPE to offer a scalable, lower cost extrusion option with potentially improved bulk mechanical properties.

Direct Extrusion↗

Accurate and Sensitive Analytical Strategy for the Determination of Antimony: Hydrogen Assisted T-Shaped Slotted Quartz Tube-Atom Trap-Flame Atomic Absorption Spectrometry

Antimony is known to have some adverse health effects on human health. Flame atomic absorption spectrometry (FAAS) is a widely used instrumental for the determination of antimony and other metals. However, it lacks the sensitivity to determine these metals at trace levels. This study was aimed at overcoming this setback by using hydrogen assisted T-shaped slotted quartz tube technique to preconcentrate and determine antimony by FAAS. All the system parameters were optimized to enhance the detection power of the system. Under the optimum experimental conditions, the limits of detection and quantification were found to be 0.75 and 2.49 µg L{sup −1}, respectively with R{sup 2} value of 0.9999. Accuracy of the developed method was validated by experimental results agreeing to the certified value of a standard reference material. Recovery studies were also carried out to determine the method’s applicability to tap and mineral water samples, and the results obtained were appreciable.

54 ENVIRONMENTAL SCIENCES↗

Stent-Assisted Coil Embolization of a Wide-Neck Aneurysm of a Persistent Primitive Hypoglossal Artery

Persistent primitive hypoglossal artery (PPHA) represents the second most common carotid-vertebrobasilar anastomosis. The association of PPHA with intracranial aneurysms is not unusual. Treatment of aneurysms located on the PPHA itself is challenging due to the increased risk of ischemic complications secondary to the hypoglossal artery often being the sole contributor of flow to the posterior circulation. We report a case of a wide-neck aneurysm in a PPHA successfully treated using a stent-assisted coil embolization technique.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Minimally Invasive Catheter Procedures to Assist Complicated Pacemaker Lead Extraction and Implantation in the Operating Room

We report on percutaneous catheter procedures in the operating room (OR) to assist complicated manual extraction or insertion of pacemaker (PM) and implantable cardioverter defibrillator leads. We retrospectively reviewed complicated PM revisions and implantations performed between 2004 and 2009 that required percutaneous catheter procedures performed in the OR. The type of interventional procedure, catheter and retrieval system used, venous access, success rates, and procedural complications were analyzed. In 41 (12 female and 29 male [mean age 62 {+-} 17 years]) of 3021 (1.4%) patients, standard manual retrieval of old leads or insertion of new leads was not achievable and thus required percutaneous catheter intervention for retrieval of misplaced leads and/or recanalisation of occluded central veins. Thirteen of 18 (72.2%) catheter-guided retrieval procedures for misplaced (right atrium [RA] or ventricle [RV; n = 3], superior vena cava [n = 2], brachiocephalic vein [n = 5], and subclavian vein [n = 3]) lead fragments in 16 patients were successful. Percutaneous catheter retrieval failed in five patients because there were extremely fixed or adhered lead fragments. Percutaneous transluminal angiography (PTA) of central veins for occlusion or high-grade stenosis was performed in 25 patients. In 22 of 25 patients (88%), recanalization of central veins was successful, thus enabling subsequent lead replacement. Major periprocedural complications were not observed. In the case of complicated manual PM lead implantation or revision, percutaneous catheter-guided extraction of misplaced lead fragments or recanalisation of central veins can be performed safely in the OR, thus enabling subsequent implantation or revision of PM systems in the majority of patients.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Stent-Assisted Coil Embolization of a Mycotic Renal Artery Aneurysm by Use of a Self-Expanding Neurointerventional Stent

Mycotic aneurysms are uncommon, especially those located in visceral arteries. We present a case of a patient with two visceral mycotic aneurysms due to bacterial endocarditis, one located in right upper pole renal artery and the second in the splenic artery. Both aneurysms were treated as endovascular embolization using microcoils. In the aneurysm located at the renal artery, the technique of stent-assisted coils embolization was preferred to avoid coils migration due to its wide neck. The stent used was the Solitaire AB, which was designed for the treatment of intracranial aneurysms and was used recently in acute stroke as a mechanical thrombectomy device. Complete embolization of the aneurysm was achieved, preserving all the arterial branches without nephrogram defects in the final angiogram.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Value of Duplex Ultrasound Assistance for Thromboaspiration and Dilation of Thrombosed Native Arterio-Venous Fistulae

Purpose: To evaluate the value of duplex ultrasound assistance during thromboaspiration of thrombosed arteriovenous fistulae for haemodialysis. Materials and Methods: We prospectively studied 54 thrombosed native fistulae (23 with total thrombosis and 31 with partial thrombosis), in which we performed manual thromboaspiration guided by ultrasonography associated with fluoroscopy. Results: The fistulae were located in the forearm (n = 39) or in the upper arm (n = 15) of 46 patients. Mean patient age was 65 years, and hypertension was the most common risk factor (74 %). Mean access age was 928 days (range 69-2,290), and most fistulae were on the left side (41 cases, 75.92 %). The success rate was 83 % in the total thrombosis group and 100 % in the partial thrombosis group. Including initial failures, the respective primary patency rates in the total thrombosis group and the partial thrombosis group were, respectively, 83 {+-} 8 % (n = 20) and 87 {+-} 6 % (n = 28) at 1 month, 39 {+-} 10 % (n = 10) and 61 {+-} 8 % (n = 20) at 6 months, and 17 {+-} 8 % (n = 5) and 26 {+-} 8 % (n = 9) at 1 year. The mean decrease of fluoroscopy time with ultrasound was 3 min (range 1-5). The mean decrease of radiation dose was 2.6 Gy cm Superscript-Two (range 0.9-4.3]. Conclusion: Ultrasound is a feasible and useful tool in the management of thrombosed native fistulae, thus decreasing radiation exposure, and has no detrimental effect on success rates.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Change in Imaging Findings on Angiography-Assisted CT During Balloon-Occluded Transcatheter Arterial Chemoembolization for Hepatocellular Carcinoma

PurposeTo evaluate changes in imaging findings on CT during hepatic arteriography (CTHA) and CT during arterial portography (CTAP) by balloon occlusion of the treated artery and their relationship with iodized oil accumulation in the tumor during balloon-occluded transcatheter arterial chemoembolization (B-TACE).MethodsBoth B-TACE and angiography-assisted CT were performed for 27 hepatocellular carcinomas. Tumor enhancement on selective CTHA with/without balloon occlusion and iodized oil accumulation after B-TACE were evaluated. Tumorous portal perfusion defect size on CTAP was compared with/without balloon occlusion. Factors influencing discrepancies between selective CTHA with/without balloon occlusion and the degree of iodized oil accumulation were investigated.ResultsAmong 27 tumors, tumor enhancement on selective CTHA changed after balloon occlusion in 14 (decreased, 11; increased, 3). In 18 tumors, there was a discrepancy between tumor enhancement on selective CTHA with balloon occlusion and the degree of accumulated iodized oil, which was higher than the tumor enhancement grade in all 18. The tumorous portal perfusion defect on CTAP significantly decreased after balloon occlusion in 18 of 20 tumors (mean decrease from 21.9 to 19.1 mm in diameter; p = 0.0001). No significant factors influenced discrepancies between selective CTHA with/without balloon occlusion. Central area tumor location, poor tumor enhancement on selective CTHA with balloon occlusion, and no decrease in the tumorous portal perfusion defect area on CTAP after balloon occlusion significantly influenced poor iodized oil accumulation in the tumor.ConclusionsTumor enhancement on selective CTHA frequently changed after balloon occlusion, which did not correspond to accumulated iodized oil in most cases.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Comparison of Balloon-Occluded Retrograde Transvenous Obliteration (BRTO) Using Ethanolamine Oleate (EO), BRTO Using Sodium Tetradecyl Sulfate (STS) Foam and Vascular Plug-Assisted Retrograde Transvenous Obliteration (PARTO)

PurposeTo compare the clinical outcomes of balloon-occluded retrograde transvenous obliteration (BRTO) using ethanolamine oleate (EO), BRTO using sodium tetradecyl sulfate (STS) foam, and vascular plug-assisted retrograde transvenous obliteration (PARTO).Materials and MethodsFrom April 2004 to February 2015, ninety-five patients underwent retrograde transvenous obliteration for gastric varices were analyzed retrospectively. BRTO with EO was performed in 49 patients, BRTO with STS foam in 25, and PARTO in 21. Among them, we obtained follow-up data in 70 patients. Recurrence of gastric varices was evaluated by follow-up endoscopy or CT. Medical records were reviewed for the clinical efficacy. Statistical analyses were performed by Kaplan–Meier method, Chi-square, Fisher’s, and Kruskal–Wallis tests.ResultsTechnical and clinical success was 94.7 %. As major complications, a hemoglobinuria and a death due to disseminated intravascular coagulation (DIC) were occurred in two patients with BRTO using EO. Recurrence occurred more frequently in PARTO group (P < 0.05). Recurrence occurred in three patients in BRTO using EO group and four patients in PARTO group with 3.2 and 32.8 % of each expected 1-year recurrence rates. There was no recurrence in BRTO using STS group. Abdominal pain occurred more frequently in BRTO using EO than BRTO using STS foam and PARTO (P < 0.05). Procedure time of PARTO was shorter than two conventional BRTOs (P < 0.05).ConclusionsBRTO using STS foam or PARTO is better than BRTO using EO for treatment of gastric varices in terms of complication or procedure time. However, PARTO showed frequent recurrence of gastric varices during the long-term follow-up rather than BRTO.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

A Retrospective Comparison of Ultrasound-Assisted Catheter-Directed Thrombolysis and Catheter-Directed Thrombolysis Alone for Treatment of Proximal Deep Vein Thrombosis

BackgroundRecent studies have suggested that catheter-directed thrombolysis (CDT) reduces development of post-thrombotic syndrome (PTS). Ultrasound-assisted CDT (USCDT) might enhance the efficiency of thrombolysis. We aimed to compare USCDT with CDT on efficacy, safety, development of PTS, and quality of life after long-term follow-up.MethodsWe describe a retrospective case series of 94 consecutive patients admitted with iliofemoral or more proximal deep vein thrombosis (DVT) to the University Hospital from 2002 to 2011, treated either with CDT or USCDT. Scheduled follow-up visits took place between April 2013 and January 2014. Venography measured the degree of residual luminal obstruction of the affected veins. Each patient completed the Short Form 36-item health survey assessment and the Venous Insufficiency Epidemiological and Economic Study-Quality of Life/Symptoms questionnaires. PTS was assessed using the Villalta scale.ResultsRisk factors of DVT were equally distributed between groups. In the USCDT group, we observed a significant decline in the duration of thrombolytic treatment (<48 h: 27 vs. 10 %), shortened hospital stay (median 6.0 days (IQR 5.0–9.0) vs. 8.0 (IQR 5.8–12.0)), and less implantation of (intravenous) stents (30 vs. 55 %). There was no difference in patency (76 vs. 79 % fully patent), prevalence of PTS (52 vs. 55 %), or quality of life between groups after long-term follow-up (median 65 months, range: 15–141).ConclusionsIn this observational study, USCDT was associated with shortened treatment duration, shorter hospital stay, and less intravenous stenting, compared to CDT alone without affecting the long-term prevalence of PTS or quality of life.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Minilaparotomy-Assisted Transmesenteric-Transjugular Intrahepatic Portosystemic Shunt: Comparison with Conventional Transjugular Approach

PurposeThis study was performed to compare the intrahepatic shunt function outcome and procedural complications of minilaparotomy-assisted transmesenteric (MAT)-transjugular intrahepatic portosystemic shunt (TIPS) placement with the conventional transjugular method.MethodsThis is a retrospective review of all patients who had a MAT or conventional TIPS procedure over a 6-year period at our institute. The primary patency rate, fluoroscopy time, technical success, major procedure-related complications, and mortality data were compared between two treatment groups.ResultsWe included 49 patients with MAT-TIPS, and 63 with conventional TIPS, with an average follow-up of 21.43 months. The primary patency rates at 6 and 12 months were 82.9 and 66.7 % in the conventional TIPS group, and 81.0 and 76.5 % in the MAT-TIPS group (p = 1.000, and 0.529), respectively. There was no significant difference in technical success rate, post-procedure portosystemic pressure gradient, fluoroscopy time, and peri-procedural mortality rate between treatment groups. Major procedural-related complications were seen more frequently among MAT-TIPS patients (p = 0.012). In the MAT-TIPS group, 5 (10.2 %) patients developed post-procedure minilaparotomy, wound-related complications, and 5 (10.2 %) developed bacterial peritonitis; whereas, none of patients with conventional TIPS had either of these complications (p = 0.014).ConclusionWhile both MAT-TIPS and conventional TIPS had similar shunt primary patency rate and technical success rate, the MAT approach was associated with a significantly higher rate of minilaparotomy-related wound complications or infectious complications. These complications maybe prevented by a change in post-procedure monitoring and therapy.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Percutaneous Radiofrequency Ablation Versus Robotic-Assisted Partial Nephrectomy for the Treatment of Small Renal Cell Carcinoma

IntroductionThe authors compared the oncologic outcomes of radiofrequency ablation (RFA) with robotic-assisted partial nephrectomy (RPN) for the treatment of T1 stage renal cell carcinoma (RCC).Materials and methodsThis was a retrospective data analysis of a high-volume single tertiary centre. Patients were treated with RFA or RPN following multidisciplinary decision making. Only histologically proven RCCs were included. Baseline demographics were collected, and PADUA scores of tumour features were calculated to standardize baseline anatomy. Peri-operative complications, kidney function and oncological outcomes were compared.ResultsSixty-three cases were included in each group. Baseline renal function was poorer in RFA, and 16/63 RFA patients had tumours in single kidneys compared to 1/63 RPN cases (p < 0.001). Length of stay was shorter in RFA (1 vs. 3 days, p < 0.0001). Post-procedure renal function decline at 30 days was significantly less in RFA [(−0.8) ± 9.6 vs. (−16.1) ± 19.5 mls/min/1.73 m{sup 2}; p < 0.0001]. More minor complications were recorded in RPN (10/63 vs. 4/63, p = 0.15), but local recurrence was numerically higher in RFA (6/63 vs. 1/63, p = 0.11). Disease-free survival (DFS) was not significantly different (adjusted HR = 0.6, 95 % Cl 0.1–3.7; p = 0.60). Increasing tumour size was an independent predictor of local recurrence (adjusted HR = 1.7; 95 % Cl 1.1–2.6 per cm; p = 0.02).ConclusionsBoth RPN and RFA offer very good oncological outcomes for the treatment of T1 RCC with low peri-operative morbidity and similar oncologic outcomes. RFA demonstrated fewer peri-operative complications and better preservation of renal function, whereas RPN had an insignificantly lower local recurrence rate. RFA should be offered alongside RPN for selected cases.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Balloon-Assisted Chemoembolization Using a Micro-Balloon Catheter Alongside a Microcatheter for a Hepatocellular Carcinoma with a Prominent Arterioportal Shunt: A Case Report

Although transcatheter arterial chemoembolization is one of the established treatments for hepatocellular carcinoma (HCC), it is difficult to treat HCCs with prominent arterioportal (AP) shunts because anticancer drugs and embolic materials migrate into the non-tumorous liver through the AP shunts and may cause liver infarction. We developed a novel method of balloon-assisted chemoembolization using a micro-balloon catheter alongside a microcatheter simultaneously inserted through a single 4.5-Fr guiding sheath, comprising proximal chemoembolization with distal arterial balloon occlusion. We applied this method to treat an HCC with a prominent distal AP shunt induced by previous proton beam therapy and achieved successful chemoembolization without non-tumorous liver infarction under temporal balloon occlusion of a distal AP shunt.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Robotically Assisted CBCT-Guided Needle Insertions: Preliminary Results in a Phantom Model

Aim: To compare robotic-assisted needle insertions performed under CBCT guidance to standard manual needle insertions. Materials and Methods: A homemade robotic prototype was used by two operators to perform robotic and manual needle insertions on a custom-made phantom. Both the operators had no experience with the prototype before starting the trial. The primary endpoint was accuracy (i.e., the minimal distance between the needle tip and the center of the target) between robotic and manual insertions. Secondary endpoints included total procedure time and operators’ radiation exposure. The Wilcoxon test was used. A p value less than 0.05 was considered statistically significant. Results: Thirty-three (17 manual, 16 robotic) needle insertions were performed. Mean accuracy for robotic insertion was 2.3 ± 0.9 mm (median 2.1; range 0.8–4.2) versus 2.3 ± 1 mm (median 2.1; range 0.7–4.4) for manual insertion (p = 0.84). Mean procedure time was 683 ± 57 s (median 670; range 611–849) for the robotic group versus 552 ± 40 s (median 548; range 486–621) for the manual group (p = 0.0002). Mean radiation exposure was 3.25 times less for the robotic insertion on comparison to manual insertion for the operator 1 (0.4 vs 1.3 µGy); and 4.15 times less for the operator 2 (1.9 vs 7.9 µGy). Conclusion: The tested robotic prototype showed accuracy comparable to that achieved with manual punctures coupled to a significant reduction of operators’ radiation exposure. Further, in vivo studies are necessary to confirm the efficiency of the system.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Robotically Assisted Sonic Therapy (RAST) for Noninvasive Hepatic Ablation in a Porcine Model: Mitigation of Body Wall Damage with a Modified Pulse Sequence

Purpose: Robotically assisted sonic therapy (RAST) is a nonthermal, noninvasive ablation method based on histotripsy. Prior animal studies have demonstrated the ability to create hepatic ablation zones at the focal point of an ultrasound therapy transducer; however, these treatments resulted in thermal damage to the body wall within the path of ultrasound energy delivery. The purpose of this study was to evaluate the efficacy and safety of a pulse sequence intended to mitigate prefocal body wall injury. Materials and Methods: Healthy swine (n = 6) underwent hepatic RAST (VortxRx software version 1.0.1.3, HistoSonics, Ann Arbor MI) in the right hepatic lobe. A 3.0 cm spherical ablation zone was prescribed for each. Following treatment, animals underwent MRI which was utilized for ablation zone measurement, evaluation of prefocal injury, and assessment of complications. Each animal was euthanized, underwent necropsy, and the tissue was processed for histopathologic analysis of the ablation zone and any other sites concerning for injury. Results: No prefocal injury was identified by MRI or necropsy in the body wall or tissues overlying the liver. Ablation zones demonstrated uniform cell destruction, were nearly spherical (sphericity index = 0.988), and corresponded closely to the prescribed size (3.0 × 3.1 × 3.4 cm, p = 0.70, 0.36, and 0.01, respectively). Ablation zones were associated with portal vein (n = 3, one occlusive) and hepatic vein thrombosis (n = 4, one occlusive); however, bile ducts remained patent within ablation zones (n = 2). Conclusions: Hepatic RAST performed with a modified ultrasound pulse sequence in a porcine model can mitigate prefocal body wall injuries while maintaining treatment efficacy. Further study of hepatic RAST appears warranted, particularly in tumor models.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

The effect of the target structure and composition on the ejection and transport of polymer molecules and carbon nanotubes in matrix-assisted pulsed laser evaporation

Matrix-assisted pulsed laser evaporation (MAPLE) is a prominent member of a broad and expanding class of laser-driven deposition techniques where a matrix of volatile molecules absorbs laser irradiation and provides the driving force for the ejection and transport of the material to be deposited. The mechanisms of MAPLE are investigated in coarse-grained molecular dynamic simulations focused on establishing the physical regimes and limits of the molecular transfer from targets with different structures and compositions. The systems considered in the simulations include dilute solutions of polymer molecules and individual carbon nanotubes (CNTs), as well as continuous networks of carbon nanotubes impregnated with solvent. The polymer molecules and nanotubes are found to be ejected only in the ablation regime and are incorporated into matrix-polymer droplets generated in the process of the explosive disintegration of the overheated matrix. The ejection and deposition of droplets explain the experimental observations of complex surface morphologies in films deposited by MAPLE. In simulations performed for MAPLE targets loaded with CNTs, the ejection of individual nanotubes, CNT bundles, and tangles with sizes comparable or even exceeding the laser penetration depth is observed. The ejected CNTs align along the flow direction in the matrix plume and tend to agglomerate into bundles at the initial stage of the ablation plume expansion. In a large-scale simulation performed for a target containing a network of interconnected CNT bundles, a large tangle of CNT bundles with the total mass of 50 MDa is separated from the continuous network and entrained with the matrix plume. No significant splitting and thinning of CNT bundles in the ejection process is observed in the simulations, suggesting that fragile structural elements or molecular agglomerates with complex secondary structures may be transferred and deposited to the substrate with the MAPLE technique.

Materials Science↗

MABAL: a Novel Deep-Learning Architecture for Machine-Assisted Bone Age Labeling

Bone age assessment (BAA) is a commonly performed diagnostic study in pediatric radiology to assess skeletal maturity. The most commonly utilized method for assessment of BAA is the Greulich and Pyle method (Pediatr Radiol 46.9:1269–1274, 2016; Arch Dis Child 81.2:172–173, 1999) atlas. The evaluation of BAA can be a tedious and time-consuming process for the radiologist. As such, several computer-assisted detection/diagnosis (CAD) methods have been proposed for automation of BAA. Classical CAD tools have traditionally relied on hard-coded algorithmic features for BAA which suffer from a variety of drawbacks. Recently, the advent and proliferation of convolutional neural networks (CNNs) has shown promise in a variety of medical imaging applications. There have been at least two published applications of using deep learning for evaluation of bone age (Med Image Anal 36:41–51, 2017; JDI 1–5, 2017). However, current implementations are limited by a combination of both architecture design and relatively small datasets. The purpose of this study is to demonstrate the benefits of a customized neural network algorithm carefully calibrated to the evaluation of bone age utilizing a relatively large institutional dataset. In doing so, this study will aim to show that advanced architectures can be successfully trained from scratch in the medical imaging domain and can generate results that outperform any existing proposed algorithm. The training data consisted of 10,289 images of different skeletal age examinations, 8909 from the hospital Picture Archiving and Communication System at our institution and 1383 from the public Digital Hand Atlas Database. The data was separated into four cohorts, one each for male and female children above the age of 8, and one each for male and female children below the age of 10. The testing set consisted of 20 radiographs of each 1-year-age cohort from 0 to 1 years to 14–15+ years, half male and half female. The testing set included left-hand radiographs done for bone age assessment, trauma evaluation without significant findings, and skeletal surveys. A 14 hidden layer-customized neural network was designed for this study. The network included several state of the art techniques including residual-style connections, inception layers, and spatial transformer layers. Data augmentation was applied to the network inputs to prevent overfitting. A linear regression output was utilized. Mean square error was used as the network loss function and mean absolute error (MAE) was utilized as the primary performance metric. MAE accuracies on the validation and test sets for young females were 0.654 and 0.561 respectively. For older females, validation and test accuracies were 0.662 and 0.497 respectively. For young males, validation and test accuracies were 0.649 and 0.585 respectively. Finally, for older males, validation and test set accuracies were 0.581 and 0.501 respectively. The female cohorts were trained for 900 epochs each and the male cohorts were trained for 600 epochs. An eightfold cross-validation set was employed for hyperparameter tuning. Test error was obtained after training on a full data set with the selected hyperparameters. Using our proposed customized neural network architecture on our large available data, we achieved an aggregate validation and test set mean absolute errors of 0.637 and 0.536 respectively. To date, this is the best published performance on utilizing deep learning for bone age assessment. Our results support our initial hypothesis that customized, purpose-built neural networks provide improved performance over networks derived from pre-trained imaging data sets. We build on that initial work by showing that the addition of state-of-the-art techniques such as residual connections and inception architecture further improves prediction accuracy. This is important because the current assumption for use of residual and/or inception architectures is that a large pre-trained network is required for successful implementation given the relatively small datasets in medical imaging. Instead we show that a small, customized architecture incorporating advanced CNN strategies can indeed be trained from scratch, yielding significant improvements in algorithm accuracy. It should be noted that for all four cohorts, testing error outperformed validation error. One reason for this is that our ground truth for our test set was obtained by averaging two pediatric radiologist reads compared to our training data for which only a single read was used. This suggests that despite relatively noisy training data, the algorithm could successfully model the variation between observers and generate estimates that are close to the expected ground truth.

62 RADIOLOGY AND NUCLEAR MEDICINE↗