Cesar A. Santana

Mining constrained association rules to predict heart disease

This work describes our experiences in discovering association rules in medical data to predict heart disease. We focus on two aspects of this work: mapping medical data to a transaction format suitable for mining association rules, and identifying useful constraints. Based on these aspects we introduce an improved algorithm to discover constrained association rules. We present an experimental section explaining several interesting discovered rules.

Quantitative (82)Rb PET/CT: development and validation of myocardial perfusion database.

The use of myocardial perfusion 82Rb PET/CT studies continues to increase but its accuracy using database quantification methods for the diagnosis of coronary artery disease (CAD) has not been established. Methods: A sex-independent normal database and criteria for abnormality for rest–stress 82Rb PET/CT myocardial perfusion imaging were developed and validated by evaluation of 281 patients (136 females: mean age ± SD, 63.3 ± 13.3 y; 145 males: mean age ± SD, 63.9 ± 12.8 y) who underwent a rest–adenosine stress 82Rb PET/CT study. These patients were divided into 3 groups: (a) healthy group: 30 patients, with <5% likelihood of CAD (low likelihood [LLK]) based on sequential Bayesian analysis; these patients were used to generate the normal distribution; (b) pilot group: 174 patients; these patients were used to determine the optimal criteria for detecting and localizing the perfusion abnormality; and (c) validation group: 76 patients (23 with LLK of CAD and 53 who underwent coronary angiography; these patients were used for prospective validation. Results: Of the 53 patients who underwent coronary angiography, 8 had <50% stenosis and 45 patients had at least one stenosis ≥50% in one major artery. Fifteen patients had single-vessel disease, 17 had double-vessel disease, and 13 had triple-vessel disease. The prospective validation shows a normalcy rate of 78% (18/23) for global CAD. The analyses by individual arteries show a normalcy rate of 96% (22/23) for the left anterior descending coronary artery, 96% for the left circumflex coronary artery (22/23), and 100% for the right coronary artery (23/23). The overall sensitivity for detection of CAD (≥50% stenosis) was 93% (42/45). The overall specificity for detection of the absence of CAD (≤50% stenosis) was 75% (6/8). Also, the positive predictive value for global CAD was 95% (42/44), the negative predictive value was 67% (6/9), and the accuracy was 91% (48/53). Conclusion: The quantitative 82Rb PET/CT database created and validated in this study is highly accurate for the detection and localization of CAD. Physicians should consider using the quantitative output of these algorithms as decision support tools to aid with image interpretation.

Adenosine stress rubidium-82 PET/computed tomography in patients with known and suspected coronary artery disease.

BackgroundIntegrated positron emission tomography/computed tomography (PET/CT) is increasingly being utilized for myocardial perfusion imaging (MPI). However, there is a potential for increased imaging artifact compared with standard PET due to the different temporal resolution of PET and CT. We reviewed the diagnostic accuracy of adenosine stress 82Rb myocardial perfusion PET/CT to detect obstructive coronary artery disease (CAD) on invasive angiography at our institution. Methods and resultsSeventy-five patients were included, 23 (13 men, mean age 55.8±11.8 years) with low likelihood of CAD and 52 (28 men, mean age 67.1±11.4 years) with intermediate to high pretest probability of disease. Coronary angiography was performed only in the latter 52 patients on average within 17 days of the MPI study. The test characteristics of PET/CT MPI were assessed using a threshold of ≥50 and ≥70% stenosis in one or more major coronary artery on invasive angiography. Dedicated software was used for registration, processing, and interpretation. Consensus interpretation of the tomographic PET slices using a 4-point scale (1=definitely normal, 2=probably normal, 3=probably abnormal, 4=definitely abnormal) was done by two readers blinded to clinical information. ResultsAll MPI studies in the 23 low likelihood patients were normal. In the remaining 52 patients using a stenosis severity ≥50%, global sensitivity and specificity, negative and positive predictive value for detection of CAD were 86, 100, 57, and 100%. Using a stenosis severity ≥70%, these values changed to 90, 83, 71, and 87%. ConclusionAdenosine stress 82Rb MPI using PET/CT with manual registration demonstrates diagnostic accuracy comparable with that of traditional PET MPI.

Normal values and prospective validation of transient ischaemic dilation index in 82Rb PET myocardial perfusion imaging.

BackgroundThe use of 82Rb positron emission tomography (PET) for the diagnosis of coronary artery disease (CAD) has increased in recent years but the role of some of the traditional parameters used in SPECT for the diagnosis of CAD, such as transient ischaemic dilation index (TID) of the left ventricle, have not been validated in PET studies. Methods and resultsWe studied 95 patients who had undergone rest/pharmacological stress 82Rb PET scans. Thirty of these patients (18 female and 12 male) who had less than 5% likelihood of CAD (LLK) based on sequential Bayesian analysis, were used to determine the normal limits of TID index in this protocol. The remaining 65 patients (33 female and 32 male) underwent coronary angiography within 15 days of the cardiac PET scan. This second group of patients was used to validate the TID normal limits determined in the first group. In LLK patients mean TID index was 1.01±0.07 and there were no significant differences between genders. The TID index upper normal limit was 1.15 and was calculated as mean+2 SD. Using this cut-off point, TID index had high specificity and PPV in the diagnosis of single vessel CAD (100% and 100% respectively) and multiple vessel CAD (93% and 85%, respectively). ConclusionOur results indicate that elevated TID index is a specific, although not sensitive marker of single and multiple vessel CAD in pharmacologically stressed 82Rb PET myocardial perfusion studies.

Prognostic performance of quantitative PET tools for stratification of patients with ischemic cardiomyopathy undergoing myocardial viability assessment.

ObjectivesThis study was performed to determine the prognostic performance of quantitative PET tools in the stratification of patients with ischemic cardiomyopathy undergoing myocardial viability assessment. MethodsWe applied four different quantitative tools to 104 consecutive patients with coronary artery disease and previous myocardial infarction who had undergone rest 82Rb/gated 18F-fluorodeoxyglucose (FDG) PET, to assess myocardial viability for potential revascularization. One of these tools was based on the FDG study alone and the other three tools assessed the extent of match/mismatch defects using FDG in comparison with a perfusion reference database. The four quantitative tools used in this research to define viability were (i) FDG alone, which calculates the percentage of left ventricular myocardium (LVM) that is above the 50% of the maximum LVM FDG counts, (ii) low flow match/mismatch, which determines the area with a 5% increase in normalized FDG counts in relation to defined resting perfusion defects as compared with a reference database, (iii) all regions match/mismatch, which computes the area with a 10% increase in normalized FDG counts in relation to the left ventricle resting perfusion distribution, and (iv) percentage max FDG match/mismatch, which defines the area with FDG uptake greater than 60% of the maximum LVM FDG counts within defined perfusion defects as determined by the reference database. The primary endpoint for this analysis was cardiac death. ResultsDuring the follow-up period (22±14 months), 19 patients (18%) died; in 17 of these the cause of death was cardiac. Using univariate analysis, none of the methods were predictive of cardiac death. Receiver operating characteristic analysis defined the optimal thresholds for the extent of myocardial viability for the four tools in the prediction of cardiac death: FDG alone=20%, low flow match/mismatch=15%, all regions match/mismatch=35%, and percentage max FDG match/mismatch=20%. A censored survival analysis using a Kaplan–Meier method showed a statistically significant difference between patients with cardiac death and those with no cardiac death using only the low flow match/mismatch (hazard ratio=0.29, P=0.01) and percentage max FDG match/mismatch criteria (hazard ratio=0.23, P=0.005) tools. ConclusionThe low flow match/mismatch and percentage max FDG match/mismatch quantitative PET tools are useful for prognostic stratification of patients with ischemic cardiomyopathy undergoing myocardial viability assessment.

Pitfalls and Artifacts in Cardiac Imaging

ECG gated myocardial perfusion single photon emission computerized tomographic (SPECT) studies, properly acquired, processed, displayed, quantified, and interpreted, provide important perfusion and function information that has been shown to have high efficacy in the diagnosis, prognosis, and management of patients with coronary artery disease (CAD). Totally automatic computer processing software packages have been developed and validated for assisting physicians in their assessment of myocardial perfusion and function. These programs help to expedite, standardize, and objectify image processing and interpretation.1–3

Correct interpretation of a myocardial perfusion SPECT study in a patient with Ebstein’s anomaly through recognition of septal artifact

ConclusionQuantitative assessment by use of polar maps and normal files increases the confidence of interpretation in most cases.4,5 However, when altered anatomy not accounted for in the normal database is present, the results can be misleading. This case illustrates the necessity for complete review of the patient’s history, rotating planar images, oblique slices, and polar maps to properly interpret a study.