FlauBERT vs. CamemBERT: Understanding patient's answers by a French medical chatbot.

In a number of circumstances, obtaining health-related information from a patient is time-consuming, whereas a chatbot interacting efficiently with that patient might help saving health care professional time and better assisting the patient. Making a chatbot understand patients' answers uses Natural Language Understanding (NLU) technology that relies on 'intent' and 'slot' predictions. Over the last few years, language models (such as BERT) pre-trained on huge amounts of data achieved state-of-the-art intent and slot predictions by connecting a neural network architecture (e.g., linear, recurrent, long short-term memory, or bidirectional long short-term memory) and fine-tuning all language model and neural network parameters end-to-end. Currently, two language models are specialized in French language: FlauBERT and CamemBERT. This study was designed to find out which combination of language model and neural network architecture was the best for intent and slot prediction by a chatbot from a French corpus of clinical cases. The comparisons showed that FlauBERT performed better than CamemBERT whatever the network architecture used and that complex architectures did not significantly improve performance vs. simple ones whatever the language model. Thus, in the medical field, the results support recommending FlauBERT with a simple linear network architecture.

Effects of dataset size and interactions on the prediction performance of logistic regression and deep learning models.

BACKGROUND AND OBJECTIVE: Machine learning and deep learning models are very powerful in predicting the presence of a disease. To achieve good predictions, those models require a certain amount of data to train on, whereas this amount i) is generally limited and difficult to obtain; and, ii) increases with the complexity of the interactions between the outcome (disease presence) and the model variables. This study compares the ways training dataset size and interactions affect the performance of those prediction models. METHODS: To compare the two influences, several datasets were simulated that differed in the number of observations and the complexity of the interactions between the variables and the outcome. A few logistic regressions and neural networks were trained on the simulated datasets and their performance evaluated by cross-validation and compared using accuracy, F1 score, and AUC metrics. RESULTS: Models trained on simulated datasets without interactions provided good results: AUC close to 0.80 with either logistic regression or neural networks. Models trained on simulated dataset with order 2 interactions led also to AUCs close to 0.80 with either logistic regression or neural networks. Models trained on simulated datasets with order 4 interactions led to AUC close to 0.80 with neural networks and 0.85 with penalized logistic regressions. Whatever the interaction order, increasing the dataset size did not significantly affect model performance, especially that of machine learning models. CONCLUSION: Machine learning models were the less influenced by the dataset size but needed interaction terms to achieve good performance, whereas deep learning models could achieve good performance without interaction terms. Conclusively, with the considered scenarios, well-specified machine learning models outperformed deep learning models.

A level set framework with a shape and motion prior for segmentation and region tracking in echocardiography.

We describe a level set formulation using both shape and motion prior, for both segmentation and region tracking in high frame rate echocardiographic image sequences. The proposed approach uses the following steps: registration of the prior shape, level set segmentation constrained through the registered shape and region tracking. Registration of the prior shape is expressed as a rigid or an affine transform problem, where the transform minimizing a global region-based criterion is sought. This criterion is based on image statistics and on the available estimated axial motion data. The segmentation step is then formulated through front propagation, constrained with the registered shape prior. The same region-based criterion is used both for the registration and the segmentation step. Region tracking is based on the motion field estimated from the interframe level set evolution. The proposed approach is applied to high frame rate echocardiographic sequences acquired in vivo. In this particular application, the prior shape is provided by a medical expert and the rigid transform is used for registration. It is shown that this approach provides consistent results in terms of segmentation and stability through the cardiac cycle. In particular, a comparison indicates that the results provided by our approach are very close to the results obtained with manual tracking performed by an expert cardiologist on a Doppler Tissue Imaging (DTI) study. These preliminary results show the ability of the method to perform region tracking and its potential for dynamic parametric imaging of the heart.

An autoregressive model-based method for contrast agent detection in ultrasound radiofrequency images.

This paper presents a spectral autoregressive method dedicated to the detection of ultrasound contrast agents (USCA) from radiofrequency (rf) data. The method is based on second-order autoregressive (AR) modeling of the rf signal. Contrast agents induce a second harmonic, which may be efficiently detected through the AR spectrum using the magnitude of the second AR spectral peak (SM2). In contrast to multipulse methods that process two or more rf frames, our method processes a single rf frame. The method is tested by numerical simulation and on in vitro data for contrast agent concentrations ranging from 10(3) to 50 x 10(3) bubbles/ml (2 x 10(-6) to 10(-4) volumic concentration) and mechanical index (MI) ranging from 0.1 to 0.36. The results show that the proposed parameter SM2 enables one to detect correctly the contrast agent, in particular at low concentration and MI (the minimum difference in SM2 between tissue and USCA is 10 dB). Furthermore, the in-vitro data demonstrates that an adapted smoothing technique reduces the variability of SM2 and provides accurate and stable segmentation of the contrast agent perfusion region.

Longitudinal strain quantitates regional right ventricular contractile function.

The assessment of contractile function of the right ventricle (RV) is an important clinical issue, but this remains difficult because of its complex anatomy and structure. We thought to investigate whether new Doppler-derived myocardial deformation indexes may quantify regional contractile RV function during varying loading conditions. In nine pigs, ultrasonic crystals were inserted longitudinally in the RV inflow and outflow tracts to assess regional contractile function. The same RV segments and the interventricular septum were imaged using apical echocardiographic views. Regional function was assessed using two parameters: 1) systolic strain (SS), representing the relative magnitude of segmental systolic shortening; and 2) its temporal derivative, peak systolic strain rate (SR), i.e., the maximal velocity of segmental shortening. Data were acquired at baseline and during partial pulmonary artery constriction (PAC) and inferior vena cava occlusion (IVCO). SS decreased significantly after PAC and IVCO in both the inflow and outflow tracts but only during IVCO in the septum. SR was less sensitive to loading variations in all segments. A significant correlation was found between SS values derived from sonomicrometry and myocardial Doppler in RV segments (r = 0.84, P < 0.001). Thus regional strain and SR provide complementary information on the heterogeneous RV contractile function and can be accurately and noninvasively quantified using Doppler myocardial imaging.

Identification of acutely ischemic myocardium using ultrasonic strain measurements. A clinical study in patients undergoing coronary angioplasty.

OBJECTIVES: The goal of this study was to investigate whether the changes in myocardial deformation measured with ultrasonic strain could accurately identify acutely ischemic myocardium during coronary angioplasty. BACKGROUND: Early identification of acute myocardial ischemia has important clinical implications. The accuracy of ultrasonic strain for the detection of acute myocardial ischemia has been validated in animal experiments but has not been investigated in the clinical setting. METHODS: In 73 patients (64 +/- 12 years), either radial or longitudinal strain values were monitored in the "at-risk" segments before, during, and early after right, circumflex, and left anterior descending coronary angioplasty. Based on the visual wall motion assessed before the angioplasty, segments were divided into normokinetic (group I) and hypo/akinetic (group II). Strain data in the "at-risk " segments were compared with values derived from the adjacent nonischemic segments and normal values in 20 controls. RESULTS: Coronary occlusion induced a marked reduction in the systolic strain both in the radial (from 49 +/- 6.9% to 23 +/- 4.6% in group I and from 21.9 +/- 11% to 11.3 +/- 8.4% in group II, p < 0.001) and longitudinal directions. Concomitantly, postsystolic strain increased (from 3.8 +/- 3.1% to 14.6 +/- 9.5% in group I, and from 4.4 +/- 3.7% to 11.3 +/- 7.8% in group II in radial direction, p < 0.001). Upon reperfusion, all deformation parameters returned to near preocclusion values. In comparison with control, baseline, and reperfusion data, the systolic and postsystolic strain parameters measured during total coronary occlusion identified acutely ischemic myocardium with a sensitivity of 86% to 95% and a specificity of 83% to 89%. CONCLUSIONS: In this model of acute ischemia, ultrasonic strain indexes differentiate acutely ischemic segments from both normal and dysfunctional myocardium. This should be a promising new approach to the bedside monitoring of acute ischemic changes in regional myocardial function.

Can changes in systolic longitudinal deformation quantify regional myocardial function after an acute infarction? An ultrasonic strain rate and strain study.

OBJECTIVES: The aim of this study was to evaluate the additional value of ultrasonic strain rate and strain to myocardial velocity in the identification and quantification of regional asynergy after an acute myocardial infarction (MI). METHODS: Forty patients (59 +/- 13 years) were investigated 3 +/- 2 days after a first infarction and compared with 14 age-matched controls with normally contracting segments (group A, n = 146). Longitudinal myocardial velocities, strain rate (SR) and strain (epsilon) were postprocessed from basal, mid, and apical segments interrogated using apical views. In a subset of patients with coronary angiograms (n = 24), myocardial segments were divided into 3 groups: normally contracting segments supplied by a normal coronary artery (group B1), normally contracting segments supplied by a diseased coronary artery (group B2), and segments with abnormal motion (group B3). Velocities were decreased in patients with myocardial infarction (MI) (P <.05 vs controls) but failed to accurately differentiate normally from abnormally contracting segments. At the opposite end, systolic SR and epsilon decreased significantly with segmental asynergy severity and could identify infarct-involved segments (group B3) with a sensitivity/specificity of 85% (systolic SR and epsilon cutoff values of -0.8 s(-1) and -13%, respectively). CONCLUSION: Strain rate and strain can better assess segmental dysfunction severity than myocardial velocities alone after an acute MI.

Myocardial function defined by strain rate and strain during alterations in inotropic states and heart rate.

For porcine myocardium, ultrasonic regional deformation parameters, systolic strain (epsilon(sys)) and peak systolic strain rate (SR(sys)), were compared with stroke volume (SV) and contractility [contractility index (CI)] measured as the ratio of end-systolic strain to end-systolic wall stress. Heart rate (HR) and contractility were varied by atrial pacing (AP = 120-180 beats/min, n = 7), incremental dobutamine infusion (DI = 2.5-20 microg. kg(-1). min(-1), n = 7), or continuous esmolol infusion (0.5 mg. kg(-1). min(-1)) + subsequent pacing (120-180 beats/min) (EI group, n = 6). Baseline SR(sys) and epsilon(sys) averaged 5.0 +/- 0.4 s(-1) and 60 +/- 4%. SR(sys) and CI increased linearly with DI (20 microg. kg(-1). min(-1); SR(sys) = 9.9 +/- 0.7 s(-1), P < 0.0001) and decreased with EI (SR(sys) = 3.4 +/- 0.1 s(-1), P < 0.01). During pacing, SR(sys) and CI remained unchanged in the AP and EI groups. During DI, epsilon(sys) and SV initially increased (5 microg. kg(-1). min(-1); epsilon(sys) = 77 +/- 6%, P < 0.01) and then progressively returned to baseline. During EI, SV and epsilon(sys) decreased (epsilon(sys) = 38 +/- 2%, P < 0.001). Pacing also decreased SV and epsilon(sys) in the AP (180 beats/min; epsilon(sys) = 36 +/- 2%, P < 0.001) and EI groups (180 beats/min; epsilon(sys) = 25 +/- 3%, P < 0.001). Thus, for normal myocardium, SR(sys) reflects regional contractile function (being relatively independent of HR), whereas epsilon(sys) reflects changes in SV.

Can strain rate and strain quantify changes in regional systolic function during dobutamine infusion, B-blockade, and atrial pacing--implications for quantitative stress echocardiography.

BACKGROUND: We investigated the ability of ultrasonic strain rate (SR) and strain (epsilon) to quantify the changes in normal myocardial function at varying inotropic states and heart rates (HR) in an attempt to determine whether these new regional function indices are potentially robust enough to quantitate stress echocardiography. METHODS AND RESULTS: Twenty closed-chest pigs underwent incremental atrial pacing (AP: 120-180/min, n = 7), dobutamine infusion (DI: 2.5-20 microg/kg/min, n = 7) or esmolol infusion with subsequent pacing (EI: 0.5 +/- 0.15 mg/kg/min with pacing 120-180/min, n = 6). Radial deformation of the left ventricular posterior wall was interrogated using the parasternal short-axis view to derive regional systolic SR and epsilon values. At baseline SR and epsilon averaged 5.0 +/- 0.4 s(-1) and 60% +/- 4%, respectively. SR remained unchanged during AP and increased linearly with DI (at 2.5 microg/kg/min = 6.2 +/- 0.3 s(-1), P <.05 vs baseline; at 20 microg/kg/min = 9.9 +/- 0.7 s(-1), P <.0001 vs baseline), whereas EI resulted in a constant decrease of 30% +/- 4% in SR (P <.05). SR and left ventricular dP/dt(MAX) correlated linearly over the induced change in inotropic states and HR (r = 0.82; P <.0001). Conversely, epsilon values decreased during AP (at 180/min = 36% +/- 2%, P <.001). During DI, epsilon initially increased at 2.5 and 5 microg/kg/min (at 5 microg/kg/min = 77% +/- 6%, P <.05) and decreased for higher doses because of increasing HR. EI resulted in a decrease of 30% +/- 4% in epsilon with a further decrease during subsequent pacing. epsilon correlated linearly with left ventricular ejection fraction (r = 0.87; P <.0001). CONCLUSION: Both SR and epsilon can quantify the changes in myocardial function during a range of inotropic challenges and over the range of physiologic HRs encountered during clinical stress echocardiography. SR may reflect regional contractile function, whereas epsilon reflects changes in ventricular geometry. This study would suggest that for quantitative stress echocardiography SR is better in quantification of changes in contractile function being relatively independent of HR.

Two-dimensional ultrasonic strain rate measurement of the human heart in vivo.

A study is presented in which the feasibility of two-dimensional strain rate estimation of the human heart in vivo has been demonstrated. To do this, ultrasonic B-mode data were captured at a high temporal resolution of 3.8 ms and processed off-line. The motion of the RF signal patterns within the two-dimensional sector image was tracked and used as the basis for strain rate estimation. Both axial and lateral motion and strain rate estimates showed a good agreement with the results obtained by more established, one-dimensional techniques.

Acute changes in systolic and diastolic events during clinical coronary angioplasty: a comparison of regional velocity, strain rate, and strain measurement.

Ultrasound-derived natural strain rate and strain are new Doppler myocardial imaging (DMI) parameters, which can measure local deformation independently of overall heart motion and thus could better characterize local contractility than DMI velocities alone. This study was undertaken to evaluate the relative benefits of regional velocity, strain rate, and strain measurements in detecting the range of acute changes in regional myocardial function in the "at-risk" zone during coronary angioplasty. Sixty-one patients (aged 63 +/- 12, 18 women) with stable angina pectoris were studied before, at the end of, and during recovery from a 60-second percutaneous transluminal coronary angioplasty (PTCA) balloon occlusion. High frame rate (147 fps) color DMI regional velocity data were derived from basal posterior (parasternal view) and mid, apical septal (apical view) "at-risk" segments as well as from the corresponding segments in healthy subjects and analyzed offline for velocity (VEL), strain rate (SR), and strain (epsilon) measurements. Coronary occlusion resulted in the reduction in VEL(SYS), SR(SYS), and epsilon(SYS) values for both radial (RCA/CX occlusion) and longitudinal data (LAD occlusion) in all segments analyzed. Velocity parameters alone failed to distinguish between baseline and occlusive measurements in the "at-risk" segments with visually abnormal baseline function. SR(SYS) and epsilon(SYS) had a higher diagnostic accuracy (sensitivity 75%, 80% and specificity 80%, 82%, respectively) than VEL(SYS) velocity alone (sensitivity 68%, specificity 65%,) for identifying acute ischemia in either baseline normal and abnormal segments. DMI-derived indexes can identify and quantify the spectrum of acute systolic and diastolic ischemic changes induced during clinical PTCA. The quantitation of regional deformation rather than motion would appear to be more appropriate in detecting and quantifying acute ischemic changes in myocardial function, especially in segments with pre-existing abnormal function.

Quantification of regional left and right ventricular radial and longitudinal function in healthy children using ultrasound-based strain rate and strain imaging.

BACKGROUND: Noninvasive assessment of left (LV) and right (RV) ventricular function in children could benefit from a technique that would characterize local myocardial deformation. Color Doppler myocardial imaging (CDMI) allows the calculation of either local longitudinal or radial Strain Rate (SR) and Strain (epsilon). To determine the clinical feasibility and reproducibility of longitudinal and radial SR and epsilon, the following study was carried out. METHODS: CDMI data were obtained from 33 healthy children (4-16 years). To quantify regional longitudinal and radial function SR and epsilon data were obtained from apical and parasternal views respectively. From the extracted SR curves, peak values for systole, early diastole, and late diastole were calculated. From the extracted epsilon curves the systolic, early and late diastolic epsilon values were calculated. RESULTS: LV longitudinal deformation were homogeneous for LV basal, mid and apical segments (peak systolic SR: -1.9 +/- 0.7 s(-1), systolic epsilon -25% +/- 7%). Longitudinal SR and epsilon values were significantly higher and heterogeneous in the RV (compared with LV walls) and were maximal in the mid part of the RV free wall (peak systolic SR: -2.8 +/- 0.7 s(-1), systolic epsilon -45% +/- 13%). The RV inferior wall showed homogeneous but lower longitudinal SR and epsilon values. The LV systolic and diastolic SR and epsilon values were higher for deformation in the radial direction compared with the longitudinal direction (radial peak systolic SR: 3.7 +/- 0.9 s(-1), radial systolic epsilon 57% +/- 11%; P <.0001). The interobserver variability for radial systolic epsilon and SR was 10.3% and 13.1%, respectively. CONCLUSION: Ultrasound-based Strain SR/epsilon imaging is a practical, reproducible clinical technique, which allows the calculation of regional longitudinal and radial deformation from both LV and RV segments. The combination of regional SR/epsilon indices and the timing of specific systolic or diastolic regional events may offer a new noninvasive approach to quantifying regional myocardial function in congenital and acquired heart disease in children.