Impact of Transfer Learning Using Local Data on Performance of a Deep Learning Model for Screening Mammography.

Purpose To investigate the issues of generalizability and replication of deep learning models by assessing performance of a screening mammography deep learning system developed at New York University (NYU) on a local Australian dataset. Materials and Methods In this retrospective study, all individuals with biopsy or surgical pathology-proven lesions and age-matched controls were identified from a South Australian public mammography screening program (January 2010 to December 2016). The primary outcome was deep learning system performance-measured with area under the receiver operating characteristic curve (AUC)-in classifying invasive breast cancer or ductal carcinoma in situ (n = 425) versus no malignancy (n = 490) or benign lesions (n = 44). The NYU system, including models without (NYU1) and with (NYU2) heatmaps, was tested in its original form, after training from scratch (without transfer learning), and after retraining with transfer learning. Results The local test set comprised 959 individuals (mean age, 62.5 years ± 8.5 [SD]; all female). The original AUCs for the NYU1 and NYU2 models were 0.83 (95% CI: 0.82, 0.84) and 0.89 (95% CI: 0.88, 0.89), respectively. When NYU1 and NYU2 were applied in their original form to the local test set, the AUCs were 0.76 (95% CI: 0.73, 0.79) and 0.84 (95% CI: 0.82, 0.87), respectively. After local training without transfer learning, the AUCs were 0.66 (95% CI: 0.62, 0.69) and 0.86 (95% CI: 0.84, 0.88). After retraining with transfer learning, the AUCs were 0.82 (95% CI: 0.80, 0.85) and 0.86 (95% CI: 0.84, 0.88). Conclusion A deep learning system developed using a U.S. dataset showed reduced performance when applied "out of the box" to an Australian dataset. Local retraining with transfer learning using available model weights improved model performance. Keywords: Screening Mammography, Convolutional Neural Network (CNN), Deep Learning Algorithms, Breast Cancer Supplemental material is available for this article. © RSNA, 2024 See also commentary by Cadrin-Chênevert in this issue.

A miniature mass spectrometer for liquid chromatography applications.

A miniature mass spectrometer capable of detecting analytes eluting from a high-performance liquid chromatography (HPLC) system is described and demonstrated for the first time. The entire instrument, including all pumps and the computer, is contained within a single enclosure that may be conveniently accommodated at the base of the HPLC stack. The microspray ion source, vacuum interface, ion guide, and quadrupole ion filter are all microengineered. These components are fabricated in batches using microelectromechanical systems (MEMS) techniques and considered to be consumables. When coupled to a standard HPLC system using an integrated passive split, the limit of detection for reserpine while scanning the full mass range is 5 ng on-column (1 pg of which is passed to the microspray). The mass range is m/z 100-800, and each spectrum is typically acquired at a rate of 1 scan per second.

Theoretical demonstration of a capacitive rotor for generation of alternating current from mechanical motion.

Innovative concepts and materials are enabling energy harvesters for slower motion, particularly for personal wearables or portable small-scale applications, hence contributing to a future sustainable economy. Here we propose a principle for a capacitive rotor device and analyze its operation. This device is based on a rotor containing many capacitors in parallel. The rotation of the rotor causes periodic capacitance changes and, when connected to a reservoir-of-charge capacitor, induces alternating current. The properties of this device depend on the lubricating liquid situated between the capacitor's electrodes, be it a highly polar liquid, organic electrolyte, or ionic liquid - we consider all these scenarios. An advantage of the capacitive rotor is its scalability. Such a lightweight device, weighing tens of grams, can be implemented in a shoe sole, generating a significant power output of the order of Watts. Scaled up, such systems can be used in portable wind or water turbines.

Comparison of ion coupling strategies for a microengineered quadrupole mass filter.

The limitations of conventional machining and assembly techniques require that designs for quadrupole mass analyzers with rod diameters less than a millimeter are not merely scale versions of larger instruments. We show how silicon planar processing techniques and microelectromechanical systems (MEMS) design concepts can be used to incorporate complex features into the construction of a miniature quadrupole mass filter chip that could not easily be achieved using other microengineering approaches. Three designs for the entrance and exit to the filter consistent with the chosen materials and techniques have been evaluated. The differences between these seemingly similar structures have a significant effect on the performance. Although one of the designs results in severe attenuation of transmission with increasing mass, the other two can be scanned to m/z = 400 without any corruption of the mass spectrum. At m/z = 219, the variation in the transmission of the three designs was found to be approximately four orders of magnitude. A maximum resolution of M/DeltaM = 87 at 10% peak height has been achieved at m/z = 219 with a filter operated at 6 MHz and constructed using rods measuring (508 +/- 5) microm in diameter.

Platelet nitric oxide responsiveness: a novel prognostic marker in acute coronary syndromes.

OBJECTIVE: Nitric oxide (NO) is critically important in the regulation of vascular tone and the inhibition of platelet aggregation. We have shown previously that patients with acute coronary syndromes (ACS) or stable angina pectoris have impaired platelet responses to NO donors when compared with normal subjects. We tested the hypotheses that platelet hyporesponsiveness to NO is a predictor of (1) cardiovascular readmission and/or death and (2) all-cause mortality in patients with ACS (unstable angina pectoris or non-Q-wave myocardial infarction). METHODS AND RESULTS: Patients (n=51) with ACS had evaluation of platelet aggregation within 24 hours of coronary care unit admission using impedance aggregometry. Patients were categorized as having "normal" (> or =32% inhibition of ADP-induced aggregation with the NO donor sodium nitroprusside; 10 micromol/L; n=18) or "impaired" (<32% inhibition of ADP-induced aggregation; n=33) NO responses. We then compared the incidence of cardiovascular readmission and death during a median of 7 years of follow-up in these 2 groups. Using a Cox proportional hazards model adjusting for age, sex, index event, postdischarge medical treatment, revascularization status, left ventricular systolic dysfunction, concurrent disease states, and cardiac risk factors, impaired NO responsiveness was associated with an increased risk of the combination of cardiovascular readmission and/or death (relative risk, 2.7; 95% CI, 1.03 to 7.10; P=0.041) and all-cause mortality (relative risk, 6.3; 95% CI, 1.09 to 36.7; P=0.033). CONCLUSIONS: Impaired platelet NO responsiveness is a novel, independent predictor of increased mortality and cardiovascular morbidity in patients with high-risk ACS.

Loss of anti-aggregatory effects of aortic valve tissue in patients with aortic stenosis.

BACKGROUND AND AIM OF THE STUDY: Patients with aortic stenosis (AS) exhibit increased platelet aggregability, and thrombus formation has been documented on calcific and severely stenosed valves. Isolated porcine and canine aortic valves (AV) release nitric oxide (NO) and prostacyclin, which exert local antithrombotic effects; to date, this has not been studied in humans. In the present study the possible interaction of AV tissue with platelet aggregation was examined, using fragments of AV obtained from patients with AS and aortic regurgitation (AR). METHODS: Fragments of AV tissue, excised from patients undergoing AV replacement, were co-incubated with blood samples obtained from normal subjects. The direct effects of valve tissue from patients with AS (n = 14) or with predominant AR (n = 13) on ADP-induced platelet aggregation and intraplatelet cGMP and cAMP content were compared. RESULTS: In whole blood, non-calcified AV fragments from AR patients inhibited platelet aggregation by 57 +/- 6% (p < 0.01); in platelet-rich plasma results were analogous. In order to determine whether this anti-aggregatory effect could be attributed to the valvular release of NO or prostacyclin, intraplatelet cGMP and cAMP formation was assessed, respectively. While there were no significant changes in cGMP content, cAMP increased by 26 +/- 4% (p < 0.02). Both, anti-aggregatory and cAMP-stimulating effects were similar to those produced by 10 nM prostaglandin E1, a prostacyclin mimetic. Fragments from stenotic valves did not inhibit aggregation and did not affect cGMP or cAMP. Furthermore, fragments from heavily calcified regions potentiated aggregation and, in some cases, induced spontaneous aggregation. CONCLUSION: Minimally calcified aortic valves (i.e., AR) and, therefore, presumably also normal valves, exert anti-aggregatory effects, most likely via prostacyclin release. AS is associated with a loss of this effect, thus potentially contributing to thrombotic risk.

Effect of perindopril on platelet nitric oxide resistance in patients with chronic heart failure secondary to ischemic left ventricular dysfunction.

In 15 patients with chronic heart failure of ischemic origin who were not previously treated with angiotensin-converting enzyme inhibitors, platelets exhibited hyperaggregability and impaired responsiveness to the antiaggregatory and cyclic guanosine monophosphate-stimulatory effects of nitric oxide donor sodium nitroprusside compared with normal subjects; this was paralleled by increased blood levels of superoxide radicals. Treatment with perindopril for 4 days significantly improved platelet responses to sodium nitroprusside; there was also a trend toward a decrease in superoxide radical levels.

Comparison of minimally invasive direct anterior versus posterior total hip arthroplasty based on inflammation and muscle damage markers.

BACKGROUND: A number of surgical approaches are utilized in total hip arthroplasty. It has been hypothesized that the anterior approach results in less muscle damage than the posterior approach. We prospectively analyzed biochemical markers of muscle damage and inflammation in patients treated with minimally invasive total hip arthroplasty with an anterior or posterior approach to provide objective evidence of the local soft-tissue injury at the time of arthroplasty. METHODS: Twenty-nine patients treated with minimally invasive total hip arthroplasty through a direct anterior approach and twenty-eight patients treated with the same procedure through a posterior approach were prospectively analyzed. Perioperative and radiographic data were collected to ensure cohorts with similar characteristics. Serum creatine kinase (CK), C-reactive protein (CRP), interleukin-6 (IL-6), interleukin-1 beta (IL-1ß), and tumor necrosis factor-alpha (TNF-a) levels were measured preoperatively, in the post-anesthesia-care unit (except for the CRP level), and on postoperative days 1 and 2. The Student t test and Fisher exact test were used to make comparisons between the two groups. Independent predictors of elevation in levels of markers of inflammation and muscle damage were determined with use of multivariate logistic regression analysis. RESULTS: The levels of the markers of inflammation were slightly decreased in the direct-anterior-approach group as compared with those in the posterior-approach group. The rise in the CK level in the posterior-approach group was 5.5 times higher than that in the anterior-approach group in the post-anesthesia-care unit (mean difference, 150.3 units/L [95% CI, 70.4 to 230.2]; p < 0.05) and nearly twice as high cumulatively (mean difference, 305.0 units/L [95% CI, -46.7 to 656.8]; p < 0.05). CONCLUSIONS: We believe that the anterior total hip arthroplasty approach used in this study caused significantly less muscle damage than did the posterior surgical approach, as indicated by serum CK levels. The clinical importance of the rise in the CK level needs to be delineated by additional clinical studies. The overall physiologic burden, as demonstrated by measurement of inflammation marker levels, appears to be similar between the anterior and posterior approaches. Objective measurement of muscle damage and inflammation markers provides an unbiased way of determining the immediate effects of surgical intervention in patients treated with total hip arthroplasty.

The deleterious effects of hyperglycemia on platelet function in diabetic patients with acute coronary syndromes mediation by superoxide production, resolution with intensive insulin administration.

OBJECTIVES: This study sought to assess the determinants of platelet nitric oxide (NO) responsiveness in diabetic patients admitted with acute coronary syndromes (ACS) and the short-term effects of aggressive glycemic control on these factors. BACKGROUND: Hyperglycemia is an independent risk factor for mortality in both diabetic patients and nondiabetic patients with ACS. The mechanism(s) underlying this observation and potential benefit from its correction remain uncertain. Although a reduction in NO bioavailability has been proposed, this remains untested in the ACS setting. METHODS: A total of 76 diabetic patients with ACS were studied. Putative correlations between admission blood sugar level (BSL), inhibition of platelet aggregation by the NO donor sodium nitroprusside (SNP), and superoxide (O2-) were assessed. Hyperglycemic patients (n = 60) were randomized to acute glycemic control with intravenous versus subcutaneous insulin, and changes in the aforementioned parameters were compared. Plasma levels of the endogenous inhibitor of NO synthase asymmetric dimethylarginine (ADMA) were also monitored. RESULTS: There was an inverse correlation between admission BSL and both platelet SNP response (p = 0.007) and ADMA levels (p = 0.045), and a positive correlation with O2- generation (p < 0.001). Intravenous insulin infusion resulted in a greater reduction (p < 0.001) in BSL, differentially improved platelet responsiveness to SNP (p = 0.049), and decreased O2- (p < 0.001) and ADMA levels (p = 0.049). CONCLUSIONS: A component of platelet dysfunction in diabetic patients with ACS is impaired responsiveness to the anti-aggregatory effects of NO, probably reflecting increased NO clearance by O2-. This phenomenon is reversed by acute aggressive glycemic control. These findings provide a further rationale for use of insulin therapy in acute myocardial infarction and suggest its extension to ACS patients.

Preservation of platelet responsiveness to nitroglycerine despite development of vascular nitrate tolerance.

AIMS: Organic nitrates, via nitric oxide (NO) release, induce vasodilatation and inhibit platelet aggregation. Development of nitrate tolerance in some vascular preparations may be associated with diminished responsiveness to NO. To date it is not known to what extent vascular tolerance to organic nitrates is associated with acquired platelet hypo-responsiveness to NO. In the current study we compared the acute and chronic effects of sustained release (SR) isosorbide 5' mono-nitrate (ISMN) and transdermal nitroglycerine (TD-NTG) on blood vessels (effects on apparent arterial stiffness) and platelets (effects on responsiveness to NO donors) in patients with stable angina pectoris (SAP). METHODS: Patients (n = 34) with SAP entered a blinded randomized crossover study of ISMN (120 mg) vs. intermittent TD-NTG (15 mg 24 h(-1)). Effects of each nitrate on pulse wave reflection (augmentation index (AIx)), platelet response to adenosine di-phosphate (ADP 1 micromol l(-1)), nitroglycerine (NTG 100 micromol l(-1)) and the non-nitrate NO donor sodium nitroprusside (SNP 10 micromol l(-1)), were measured pre-dose, 4 and 8 h post dose, on three occasions: 1) at the end of a pre-nitrate phase, 2) after dosing for 7 days and 3) following 14 days of full dose therapy with either nitrate. RESULTS: Acutely, both ISMN and TD-NTG markedly reduced AIx. After 14 days, these effects were significantly attenuated (ANOVA, P = 0.018) but not abolished, indicating development of nitrate tolerance. Neither nitrate preparation affected ADP (1 micromol l(-1))-induced platelet aggregation. Platelet responsiveness to NTG (100 micromol l(-1)) and SNP (10 micromol l(-1)) was not diminished during chronic nitrate therapy, and there was no evidence of 'rebound' hyper-aggregability during 'nitrate-free' periods. CONCLUSIONS: Chronic therapy with either ISMN or TD-NTG is associated with development of vascular tolerance. Despite the induction of vascular tolerance, platelet responsiveness to NTG and SNP remains unaffected. Therefore, development of vascular tolerance is unlikely to compromise the anti-aggregatory effects of organic nitrates, or those of endogenous NO.