FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W47153544> ?p ?o ?g. }

• W47153544 abstract "A Bayesian Belief Network is employed to interpret the perfusion-ventilation lung scans along with correlated chest x-rays in order to assist in diagnosis of pulmonary embolism. A rule base is applied to interpret the images for the possibility of pulmonary embolism. The rule-base is formulated using the modified PIOPED criteria and probabilities are assigned to different hypotheses based on the patient data collected by the PIOPED investigators. These form the prior probabilities for the hypothesis that pulmonary embolism is present in a patient. Based on these probabilities, an inference is drawn in terms of probability value regarding the degree of pulmonary embolism in a given patient. Testing results indicated that the Bayesian Belief Network was able to implement the PIOPED criteria in interpreting the lung scans for diagnosis of pulmonary embolism. The proposed probabilistic reasoning system aims to reduce interobserver variability in interpretation of lung scans and assist experienced as well as inexperienced observers in drawing an inference regarding the presence of pulmonary embolism in a given patient. INTRODUCTION Pulmonary embolism (PE) has been and continues to be a major health problem with considerable controversy surrounding the diagnostic approaches and their interpretation. It has been estimated that over 50,000 people die annually from this disease in the United States without being diagnosed [NIH Online Statement, 1999]. The high costs associated with the definitive diagnostic procedure of pulmonary angiography and the nonspecificity of clinical signs and symptoms contribute greatly to diagnostic problems of PE. Most pulmonary medicine physicians and angiographers agree that a perfusion/ventilation study showing mismatched abnormalities is the most reliable non-intrusive procedure to diagnose PE [Mettler et. al., 1991]. Pulmonary perfusion imaging is based on the principle of capillary blockade. Particles larger than the size of the smallest capillaries are trapped in the first capillary bed they reach after peripheral intravenous injection. Labeled particles are injected intravenously and trapped in the capillary bed of the lungs. The distribution of these particles in the lung capillaries gives a true reflection of the distribution of the pulmonary artery blood flow in the lungs. Tc-99m, Macro Aggrevated Albumin, is mostly used in perfusion scanning. Ventilation images describe the regional patterns of washin and washout of materials inspired in the lungs. Radioactive substances (Xe-133) follow the behavior of ventilation tracks and segmental ventilation of the lungs. The perfusion scans and the ventilation scans are complementary of one another and present the entire picture of blood flow as well as air flow within the lungs. Therefore in case of PE, will be observed on the perfusion scans but the ventilation scans will not have the same defects. A mismatch is required to show presence of PE [Biello et. al., 1979]. In case the overlap, it can be interpreted that possibility of PE is low and perfusion exist due to causes other than PE. The proposed approach employs Bayesian Belief Networks (BBN) [Heckerman et. al., 1995] to interpret lung scans (perfusion, ventilation and chest x-rays) in order to assist in diagnosis of PE. A rule base is applied to interpret the images for the possibility of pulmonary embolism. The rule base is derived using the modified PIOPED criteria and probabilities are assigned to different hypotheses based on the patient data collected by the PIOPED investigators [1990]. These form the prior probabilities for the hypothesis that pulmonary embolism is present in a patient. Based on these probabilities, an inference is drawn in terms of probability value regarding the degree of pulmonary embolism in a given patient. The probabilistic result is the prominent point of using Bayesian Belief Networks over Artificial Neural Network classifiers, Fuzzy Inference Networks or Expert Systems. This is because the result resembles the human observer interpretation more closely. Furthermore, the network is able to employ a learning algorithm that would modify the existing prior probabilities to take into consideration the data from the new patient. The proposed probabilistic reasoning system aims to reduce inter-observer variability in interpretation of lung scans and assist experienced as well as inexperienced observers in drawing an inference regarding the presence of PE in a given patient. BAYESIAN BELIEF NETWORK FOR DIAGNOSIS OF PULMONARY EMBOLISM The objective of Bayesian Belief Network (BBN) is to classify the possibility of pulmonary embolism in the lung scans of a patient into one of five classes: high probability, intermediate probability, low probability, very low probability or probability (normal). These form the hypothesis for the network. The decision rules are based on the PIOPED criteria, which is presented in Figure 1. The criteria analyzes the perfusion scans, ventilation scans and the chest x-rays to arrive at a probabilistic conclusion. Based on the number of observed in the perfusion scans and matches with the ventilation scans and x-rays, an inference is drawn regarding the probability of existence of pulmonary embolism. The final interpretation of the BBN is the probability that a given patient has pulmonary embolism. Figure 1. PIOPED Criteria High Probability • Greater than 2 large segmental perfusion without corresponding ventilation or roentgenographic abnormalities or substantially larger than either matching ventilation or chest roentgenogram abnormality. • Greater than 2 moderate segmental perfusion without corresponding ventilation or roentgenographic abnormalities and 1 large mismatched segmental defect. • Greater than 4 moderate segmental perfusion without ventilation or chest roentgenogram abnormalities. Intermediate probability (indeterminate) • Not falling into normal, very-low, low, or high-probability categories. • Borderline high or borderline low. • Difficult to categorize as low or high. Low probability • Nonsegmental perfusion (e.g., very small effusion causing blunting of the costophrenic angle, cardiomegaly, enlarged aorta, hila, and mediastinum, and elevated diaphragm) • Single moderate mismatched segmental perfusion defect with normal chest roentgenogram. • Any perfusion defect with a substantially larger chest roentgenogram abnormality. • or moderate segmental perfusion involving more than 4 segments in 1 lung and more than 3 segments in 1 lung region with matching ventilation either equal to or larger in size and chest roentgenogram either normal or with abnormalities substantially smaller than perfusion defects. • Greater than 3 small segmental perfusion with a normal chest roentgenogram. Very low probability • Less than or equal to 3 small segmental perfusion with a normal chest roentgenogram. Normal • No perfusion defects. Perfusion outlines exactly the shape of the lungs as seen on the chest roentgenogram (hila and aortic impressions may be seen, chest roentgenogram and/or ventilation study may be abnormal). The probability-based diagnosis is based on the findings of the in perfusion scans, ventilation scans and the chest x-rays. Number of segmental needs to be determined per the specs in the PIOPED criteria. The are further classified as large, moderate and small. A defect is large if it occupies more than 75 % of a lung segment. It is classified as moderate if it occupies between 25 % to 75 % of the segment and the defect is small if it occupies less than 25 % of the segment. Another significant variable is the number and type of matches among in ventilation and perfusion scans and the correlating chest x-rays. The BBN for diagnosis of PE base on PIOPED criteria is presented in Figure 2. Figure 2. Bayesian Belief Network for Pulmonary Embolism Diagnosis The foremost node is labeled “Defects Present?”. This assigns a probability value of 0.5 to the state that are present and 0.5 to the state that are absent. This is an initial arbitrary value that would be subjected to a learning process and would change with addition of new patient data. This serves as a parent node to 4 nodes: Large Defects, Defects, and of present. The node labeled as Large has three cases to consider. These are the probability that greater than or equal to two large (GTE2LD) and the probability that less than two large (LT2LD) will be present as well as the “no defects case (ND). The no defects case serves as a path for probability propagation in the event there are perfusion present. are mainly responsible for determining the probability of PE on higher side. These states are assigned the following probabilities. DEFECTS GTE2LD LT2LD ND Present 0.5 0.5 0.0 Absent 0.0 0.0 1.0 The node labeled as has also three cases. They are the probability of greater than or equal to four moderate sized (GTE4MD), less than four moderate sized (LT4MD) and (ND). The probability assignment table is as shown below. DEFECTS GTE4MD LT4MD ND Present 0.5 0.5 0.0 Absent 0.0 0.0 1.0 The node also groups three cases. They are the probability of greater than or equal to three small sized (GTE3SD), less than three small sized (LT3SD) and (ND). The probability assignment table is as shown below. Probability of PE for All Classes Defects Moderate Defects Small Defects Type of Match Defects Present? X-ray Normal?" @default.
• W47153544 created "2016-06-24" @default.
• W47153544 creator A5017049098 @default.
• W47153544 creator A5018680547 @default.
• W47153544 creator A5023678431 @default.
• W47153544 creator A5036822465 @default.
• W47153544 creator A5054586621 @default.
• W47153544 creator A5089093952 @default.
• W47153544 date "2003-01-01" @default.
• W47153544 modified "2023-09-27" @default.
• W47153544 title "BAYESIAN BELIEF NETWORKS AS A DIAGNOSTIC TOOL FOR PULMONARY EMBOLISM" @default.
• W47153544 cites W2052705340 @default.
• W47153544 cites W2170112109 @default.
• W47153544 cites W2950634275 @default.
• W47153544 cites W74544974 @default.
• W47153544 hasPublicationYear "2003" @default.
• W47153544 type Work @default.
• W47153544 sameAs 47153544 @default.
• W47153544 citedByCount "2" @default.
• W47153544 crossrefType "journal-article" @default.
• W47153544 hasAuthorship W47153544A5017049098 @default.
• W47153544 hasAuthorship W47153544A5018680547 @default.
• W47153544 hasAuthorship W47153544A5023678431 @default.
• W47153544 hasAuthorship W47153544A5036822465 @default.
• W47153544 hasAuthorship W47153544A5054586621 @default.
• W47153544 hasAuthorship W47153544A5089093952 @default.
• W47153544 hasConcept C107673813 @default.
• W47153544 hasConcept C126322002 @default.
• W47153544 hasConcept C154945302 @default.
• W47153544 hasConcept C2776265017 @default.
• W47153544 hasConcept C33724603 @default.
• W47153544 hasConcept C41008148 @default.
• W47153544 hasConcept C71924100 @default.
• W47153544 hasConceptScore W47153544C107673813 @default.
• W47153544 hasConceptScore W47153544C126322002 @default.
• W47153544 hasConceptScore W47153544C154945302 @default.
• W47153544 hasConceptScore W47153544C2776265017 @default.
• W47153544 hasConceptScore W47153544C33724603 @default.
• W47153544 hasConceptScore W47153544C41008148 @default.
• W47153544 hasConceptScore W47153544C71924100 @default.
• W47153544 hasLocation W471535441 @default.
• W47153544 hasOpenAccess W47153544 @default.
• W47153544 hasPrimaryLocation W471535441 @default.
• W47153544 hasRelatedWork W2067743442 @default.
• W47153544 isParatext "false" @default.
• W47153544 isRetracted "false" @default.
• W47153544 magId "47153544" @default.
• W47153544 workType "article" @default.