Michael P. Riley

Weighted finite-state transducers in speech recognition

We survey the use of weighted finite-state transducers (WFSTs) in speech recognition. We show that WFSTs provide a common and natural representation for hidden Markov models (HMMs), context-dependency, pronunciation dictionaries, grammars, and alternative recognition outputs. Furthermore, general transducer operations combine these representations flexibly and efficiently. Weighted determinization and minimization algorithms optimize their time and space requirements, and a weight pushing algorithm distributes the weights along the paths of a weighted transducer optimally for speech recognition. As an example, we describe a North American Business News (NAB) recognition system built using these techniques that combines the HMMs, full cross-word triphones, a lexicon of 40 000 words, and a large trigram grammar into a single weighted transducer that is only somewhat larger than the trigram word grammar and that runs NAB in real-time on a very simple decoder. In another example, we show that the same techniques can be used to optimize lattices for second-pass recognition. In a third example, we show how general automata operations can be used to assemble lattices from different recognizers to improve recognition performance.

Electroanatomic Substrate and Ablation Outcome for Suspected Epicardial Ventricular Tachycardia in Left Ventricular Nonischemic Cardiomyopathy

OBJECTIVESnThe aim of the study was to define the epicardial substrate and ablation outcome in patients with left ventricular nonischemic cardiomyopathy (NICM) and suspected epicardial ventricular tachycardia (VT).nnnBACKGROUNDnVentricular tachycardia in NICM often originates from the epicardium.nnnMETHODSnTwenty-two patients with NICM underwent detailed endocardial and epicardial bipolar voltage maps and VT ablation for suspected epicardial VT. Eight patients with normal hearts and idiopathic VT served to define normal epicardial electrograms. Low-voltage regions were also assessed for wide (>80 ms), split, or late electrograms.nnnRESULTSnNormal epicardial bipolar voltage was identified as >1.0 mV on the basis of the reference population. Confluent low-voltage areas were present in 18 epicardial (82%) and 12 endocardial (54%) maps and were typically over basal lateral LV. In the 18 patients with epicardial VT on the basis of activation/pacemapping, the mean epicardial area was greater than the endocardial low-voltage area (55.3 +/- 33.5 cm(2) vs. 22.9 +/- 32.4 cm(2), p < 0.01). Epicardial low-voltage areas showed 49.7% wide (>80 ms), split, and/or late electrograms rarely seen in the reference patients (2.3%). During follow-up of 18 +/- 7 months, ablation resulted in VT elimination in 15 of 21 patients (71%) including 14 of 18 patients (78%) with epicardial VT.nnnCONCLUSIONSnIn patients with NICM and VT of epicardial origin, the substrate is characterized by areas of basal LV epicardial > endocardial bipolar low voltage. The electrograms in these areas are not only small (<1.0 mV) but wide (>80 ms), split, and/or late, and help identify the substrate targeted for successful ablation.

Endocardial unipolar voltage mapping to detect epicardial ventricular tachycardia substrate in patients with nonischemic left ventricular cardiomyopathy

Background—Patients with nonischemic left ventricular cardiomyopathy (LVCM) and ventricular tachycardia (VT) have complex 3-dimensional substrate with variable involvement of the endocardium (ENDO) and epicardium (EPI). The purpose of this study was to determine whether ENDO unipolar (UNI) mapping with a larger electric field of view could identify EPI low bipolar (BIP) voltage regions in patients with LVCM undergoing VT ablation. Methods and Results—The reference value for normal ENDO unipolar voltage was determined from 6 patients without structural heart disease. Consecutive patients undergoing VT ablation over an 8-year period with detailed (>100 points) LV ENDO and EPI mapping and normal LV ENDO BIP voltage were identified. From this cohort, we compared patients with structurally normal hearts and normal EPI BIP voltage (EPI−, group 1) with patients with LVCM and low LV EPI BIP voltage regions present (EPI+, group 2). Confluent regions of ENDO UNI and EPI BIP low voltage (>2 cm2) were measured. The normal signal amplitude was >8.27 mV for LV ENDO UNI electrograms. Detailed LV ENDO-EPI maps in 5 EPI− patients were compared with 11 EPI+ patients. Confluent ENDO UNI low-voltage regions were seen in 9 of 11 (82%) of the EPI+ (group 2) patients compared with none of 5 EPI− (group 1) patients (P<0.001). In all 9 patients with ENDO UNI low voltage, the ENDO UNI low-voltage regions were directly opposite to an area of EPI BIP low voltage (61% ENDO UNI-EPI BIP low-voltage area overlap). Conclusions—EPI arrhythmia substrate can be reliably identified in most patients with LVCM using ENDO UNI voltage mapping in the absence of ENDO BIP abnormalities.

Electrocardiographic and electrophysiologic features of ventricular arrhythmias originating from the right/left coronary cusp commissure

BACKGROUNDnVentricular arrhythmias are known to originate from the aortic sinus of Valsalva.nnnOBJECTIVEnThe purpose of this study was to identify the characteristics associated with ventricular arrhythmias originating from the right coronary cusp-left coronary cusp (RCC-LCC) commissure.nnnMETHODSnThirty-seven consecutive patients with ventricular arrhythmias originating from the aortic cusp region were studied. Intracardiac echocardiography and electroanatomic mapping were used to define coronary cusp anatomy and catheter position. Ventricular arrhythmias from the RCC-LCC commissure were compared with ventricular arrhythmias originating from other sites in the aortic cusp region.nnnRESULTSnNineteen (51%) ventricular arrhythmias had an anatomic origin at the RCC-LCC commissure. Eighteen ventricular arrhythmias originated from other aortic cusp sites (4 right cusp, 7 left cusp, 3 left ventricular endocardium, 4 left ventricular epicardium anterior to aortic valve). A QS morphology in lead V(1) with notching on the downward deflection was present in 15 of 19 ventricular arrhythmias originating from the RCC-LCC commissure compared to 2 of 18 ventricular arrhythmias from other aortic cusp sites (P <.01). At the site of earliest activation, 13 of 19 patients with RCC-LCC ventricular arrhythmias had late potentials in sinus rhythm compared to 1 of 18 ventricular arrhythmias from other aortic cusp sites (P <.01). The site of successful ablation was confirmed to be above the aortic valve plane in 15 (79%) of 19 patients with RCC-LCC ventricular arrhythmias.nnnCONCLUSIONnRCC-LCC aortic cusp ventricular arrhythmias are common and have a QS morphology in lead V(1) with notching on the downward deflection with precordial transition at lead V(3). In the majority of cases, the site of successful ablation has late potentials in sinus rhythm.

Characterization of the phrenic nerve course within the epicardial substrate of patients with nonischemic cardiomyopathy and ventricular tachycardia.

BACKGROUNDnPatients with nonischemic cardiomyopathy and ventricular tachycardia (VT) often have low-voltage areas in the lateral left ventricular (LV) epicardium that serve as the VT substrate. The course of the left phrenic nerve in this region may pose a challenge to successful and safe ablation.nnnOBJECTIVEnThe purpose of this study was to delineate the left phrenic nerve course in patients with nonischemic cardiomyopathy and suspected epicardial VT and to characterize its relationship with the VT substrate.nnnMETHODSnIn 10 patients with nonischemic cardiomyopathy undergoing epicardial VT mapping and ablation, the course of the phrenic nerve was defined by pacing. The extent of epicardial LV low-voltage areas (<1.0 mV) was characterized by electroanatomic voltage mapping.nnnRESULTSnEight of 10 patients had low-voltage areas involving the lateral epicardial LV, and 7 of these 8 patients had sites of phrenic capture within these areas. Ablation was limited due to location of the phrenic nerve in two patients. In one of these patients, a balloon catheter was successfully used to mechanically protect the phrenic nerve during ablation. In the other five patients, adjacent ablation sites were targeted at which no phrenic capture with high-output pacing was demonstrated prior to ablation. In all patients undergoing ablation, the targeted VT became noninducible, and no patient demonstrated phrenic nerve injury.nnnCONCLUSIONnIn most patients with nonischemic cardiomyopathy undergoing epicardial VT ablation, the phrenic nerve courses through a lateral LV low-voltage area in proximity to potential sites for ablation. Strategies to identify and protect the phrenic nerve are important.

Localization of Atrial Fibrillation Triggers in Patients Undergoing Pulmonary Vein Isolation Importance of the Carina Region

OBJECTIVESnThis study sought to identify the origin within the pulmonary vein (PV) of reproducible atrial fibrillation (AF) triggers.nnnBACKGROUNDnTriggers for AF frequently originate from PVs. However, a systematic evaluation of the location of origin within the PV orifice and associated techniques for eliciting triggers has not been performed.nnnMETHODSnSpontaneous triggers and those provoked with isoproterenol (up to 20 microg/min) and/or cardioversion in 45 patients with AF were identified using multipolar catheter recordings. In identifying origin, PVs were divided into 17 equal segments from ipsilateral PVs with carina zone (CZ) (7 segments between the PVs) and 10 noncarina zone (NCZ) segments.nnnRESULTSnSixty-three reproducible triggers were noted in 37 of the 45 (82%) patients with 57 from PV and 6 (10%) from non-PV sites. Although triggers were identified from 26 of 34 distinct PV segments, most PV triggers (36, 63%) originated from CZ segments (p < 0.05) from both right (17 triggers) and left (19 triggers) PVs. The CZ triggers were more often spontaneous (11 of 36 in CZ vs. 2 of 21 in NCZ; p < 0.05) or elicited with CV (17 of 36 in CZ vs. 6 of 21 in NCZ; p < 0.05). In contrast, NCZ triggers were more likely to require isoproterenol to be provoked (13 of 21 [62%] vs. 8 of 36 [22%], p < 0.05).nnnCONCLUSIONSnReproducible spontaneous and provoked PV triggers initiating AF can be observed in most patients undergoing AF ablation. These triggers most commonly originate from the carina region of both right and left PVs. Noncarina PV triggers more commonly require provocation with isoproterenol infusion.

Efficacy and Risk of Atrial Fibrillation Ablation Before 45 Years of Age

Background—Young patients with atrial fibrillation (AF) tend to be more symptomatic and less willing to take long-term medications, yet catheter ablation remains recommended as second-line therapy for AF regardless of age. This study seeks to characterize the effectiveness and risk of AF ablation in the young. Methods and Results—Consecutive (n=1548) patients who underwent 2038 AF ablation procedures were included. Major procedural complications and efficacy were analyzed on the basis of age at the initial procedure: <45 years (group 1), 45 to 54 years (group 2), 55 to 64 years (group 3), and ≥65 years (group 4). AF control was defined as no or rare AF on or off antiarrhythmic drugs. The primary outcome of AF control was similar in all groups; it was achieved in 87% in group 1, 88% in group 2, 88% in group 3, and 82% in group 4 (P=0.06). However, more group 1 patients demonstrated freedom from AF off antiarrhythmic drugs (76%) compared with group 2 at 68%, group 3 at 65%, and group 4 at 53% (P<0.001). There were no major complications in group 1, 10 (1.7%) in group 2, 14 (1.4%) in group 3, and 10 (2.6%) in group 4 (P=0.01). Conclusions—In patients younger than 45 years, there is a lower major complication rate and a comparable efficacy rate, with a greater chance of being AF free without antiarrhythmic drugs. These findings suggest that it may be appropriate to consider ablative therapy as first-line therapy in this age group.

Noninvasive Programmed Ventricular Stimulation Early After Ventricular Tachycardia Ablation to Predict Risk of Late Recurrence

OBJECTIVESnThe goal of this study was to evaluate the ability of noninvasive programmed stimulation (NIPS) after ventricular tachycardia (VT) ablation to identify patients at high risk of recurrence.nnnBACKGROUNDnOptimal endpoints for VT ablation are not well defined.nnnMETHODSnOf 200 consecutive patients with VT and structural heart disease undergoing ablation, 11 had clinical VT inducible at the end of ablation and 11 recurred spontaneously. Of the remaining 178 patients, 132 underwent NIPS through their implantable cardioverter-defibrillator 3.1 ± 2.1 days after ablation. At 2 drive cycle lengths, single, double, and triple right ventricular extrastimuli were delivered to refractoriness. Clinical VT was defined by comparison with 12-lead electrocardiograms and stored implantable cardioverter-defibrillator electrograms from spontaneous VT episodes. Patients were followed for 1 year.nnnRESULTSnFifty-nine patients (44.7%) had no VT inducible at NIPS; 49 (37.1%) had inducible nonclinical VT only; and 24 (18.2%) had inducible clinical VT. Patients with inducible clinical VT at NIPS had markedly decreased 1-year VT-free survival compared to those in whom no VT was inducible (<30% vs. >80%; p = 0.001), including 33% recurring with VT storm. Patients with inducible nonclinical VT only, had intermediate 1-year VT-free survival (65%).nnnCONCLUSIONSnWhen patients with VT and structural heart disease have no VT or nonclinical VT only inducible at the end of ablation or their condition is too unstable to undergo final programmed stimulation, NIPS should be considered in the following days to further define risk of recurrence. If clinical VT is inducible at NIPS, repeat ablation may be considered because recurrence over the following year is high.

Assessing Epicardial Substrate Using Intracardiac Echocardiography During VT Ablation

Background— Intracardiac echocardiography (ICE) has played a limited role in defining the substrate for ventricular tachycardia (VT). The purpose of this study was to assess whether ICE could identify abnormal epicardial substrate in patients with nonischemic cardiomyopathy (NICM) and VT. Methods and Results— We studied 18 patients with NICM and recurrent VT who had abnormal echogenicity identified on ICE imaging. Detailed left ventricular (LV) endocardial and epicardial electroanatomic mapping was performed in all patients. Low-voltage areas (<1.0 mV) in the epicardium were analyzed. ICE imaging in the NICM group was compared to a control group of 30 patients with structurally normal hearts who underwent ICE imaging for other ablation procedures. In 18 patients (age, 53±13 years; 17 men) with NICM (ejection fraction, 37±13%), increased echogenicity was identified in the lateral LV by ICE imaging. LV endocardial electroanatomic mapping identified normal voltage in 9 patients and at least 1 confluent low-voltage area (6.6 cm2; minimum-maximum, 2.1–31.7 cm2) in 9 patients (5 posterolateral LV, 4 perivalvular LV). Detailed epicardial mapping revealed areas of low voltage (39 cm2; minimum-maximum, 18.5–96.3 cm2) and abnormal, fractionated electrograms in all 18 patients (15 posterolateral LV, 3 lateral LV). In all patients, the epicardial scar identified by electroanatomic mapping correlated with the echogenic area identified on ICE imaging. ICE imaging identified no areas of increased echogenicity in the control group. Conclusions— ICE imaging identified increased echogenicity in the lateral wall of the LV that correlated to abnormal epicardial substrate. These findings suggest that ICE imaging may be useful to identify epicardial substrate in NICM.

Intracardiac Echocardiographic Diagnosis of Thrombus Formation in the Left Atrial Appendage: A Complementary Role to Transesophageal Echocardiography

We sought to develop and validate an intracardiac echocardiography (ICE) imaging strategy for evaluation of left atrial (LA) appendage (LAA) anatomy and function to clarify equivocal findings of LAA thrombus with transesophageal echocardiography (TEE).