This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
An expert system which is utilized for diagnosis and providing therapy program after lung transplantation has been developed and evaluated by domain experts. The system captures a total of 21 diagnoses encompassing rejection, pulmonary infection and some diseases of gastrointestinal origin. The result of the system are categorized as hypotheses and evaluated based on the score, ranked by their competence in explaining the patient findings. A hypothesis is accepted as the patient's disease if it is ranked highest. The therapy knowledge is captured in the form of rules. The results demonstrate the feasibility of the expert system in diagnosing and delivering therapy program for patients who have undergone lung transplantation.
Lung transplant have been available as therapeutic option for patients with end-stage lung diseases since early 1980. Lung transplantation is for patient who experience end-stage fibrotic lung disease, pulmonary hypertension with reversible right ventricular function, chronic obstructive pulmonary disease, pulmonary vascular disease, primary pulmonary hypertension, Eisenmenger's Syndrome with congenital heart defect and end stage parenchymal lung disease with cardiac dysfunction.
Rejection of the transplanted lung is an ever-present threat. Acute rejections commonly occur during the first 3 months after transplantation. Management of lung allograft rejection consists of augmentation of immune suppressive drugs and is a successful treatment for rejection once it develops. Example of occurrence of rejection is the progressive obstructive airways diseases due to late development of bronchiolitis obliterans. Once the disease is ascertained, response to medication is poor and many of the patients succumb to either bronchiolitis obliterans or therapeutic complications. Therefore, recognizing the rejection early is crucial to treat it as soon as possible. Infection is so serious that it causes 75% of death in lung transplant patients.
Expert system is a computer based program that represents and reasons with knowledge of some specialized subject with the hope of solving problems or rendering advice. In an expert system, reasoning is done through the representation of human knowledge (e.g rules, objects), called the knowledge base. The code that performs the reasoning is independent of the representation of human knowledge. It is known as inference engine. The problem is normally solved using heuristic or approximate method, which unlike algorithmic solutions are not always guaranteed to succeed. Hence, the solutions are often specified with a measure of belief, or certainty. Each expert system are designed for a diverse range of domain including engineering, medical and etc. to perform task such as medical diagnosis and treatment of bacterial infections. One of the early expert system introduced is INTERNIST. INTERNIST is a program setup to model the actual steps of a clinician's diagnostic reasoning. The initial setting up of hypothesis is data driven and this may trigger some conjecture. The subsequent gathering of new data is modeled in a way to support or refute the hypothesis, based on stereotypical or schematic descriptions of the manifestation of each disease. INTERNIST is a system that is able to arrive to a set of mutually exclusive disease hypotheses that accounts all findings. MYCIN is another expert system which is mainly used in prescribing therapy program to counter bacterial infection. In this paper, an expert system is used both for diagnosis and therapy of conditions commonly occur in the first 2 years after lung transplantation. The conditions include rejection, infections of gastrointestinal and pulmonary origin, colitis, pulmonary embolism, diverticulitis, peptic ulcer disease and etc.
The expert system resides in within an extensive developed system for generating synthetic temporal clinical cases related to lung transplant pathology. Synthetic temporal clinical cases capture information about the clinical course, encounter and therapy of a patient.
Synthetic cases are useful for assisting the doctor in decision making and delivering individualized instruction and testing the expert systems. The case generation process begins with initializing the attributes of the patient coherent with the end-diagnosis of interest. The chronological occurrence of the disease events is modeled right after the patient has undergone a typical lung transplant operation. The characteristic of the patient are updated consistently with the disease and therapy events which affect the patient at a particular point. As each disease manifests itself in the form of signs and symptoms, a physical exam and appropriate test are scheduled. As each test result is returned, the values are assessed for deviation from baseline of the patient.
The expert system can generate one or more or even zero diagnoses. Based on the diagnoses, the therapy expert will provide a treatment plan for the patient. The simulation resumes, and the value of the attribute is the result of the interaction of the prevalent disease and the on-going therapy. The simulation will continue repetitively between diagnostic events and test, until the simulation yields the desired end-diagnosis. The entire patient course is stored in specially designed database tables for reference.
The initial rule set was based on a set of rule or premises used with the clinical transplant program which was developed with the clinical expert for this study. Further discussions with system to certain diseases led to a substantial enrich of the therapy rule base. In general basis, a positive finding of a case is not adequate in treating the diseases. The diseases have to be proved to be existed by presenting appropriate symptoms. Subsequent tests are scheduled based on the hypothesis of the system to further verify the result of the system. Tests are conducted or planned based on the suspicion of diseases. The remainder of the knowledge acquisition is repetitive to narrow down to a specific disease.
The knowledge or database of the system was begun by interviewing human expert about the different diagnoses and inference logic. It is crucial to ensure the concepts used in reasoning by human expert are paralleled to the INTERNIST expert program. Therefore, INTERNIST program rule or model setup is completely adopted from our domain. The diagnostic expert system records the common diseases that are experienced by the patient for the first 2 years. The result of the system recorded principally to the pulmonary system and the rest related to the gastrointestinal system. Two types of findings are represented; one is the gradation (example mild, moderate and severe), and positive/negative type (Tests Result). Human expert will assign values to the three attributes (evoking strength, frequency, lab detail, sign and symptom). Evoking strength estimates the subjective likelihood of the finding. The likelihood is measured 0-5 scale.
The inference engines begin by examining findings based on the inputs of the physicians and generate disease hypothesis. The steps are repeated until all hypotheses have been examined. Each verified hypothesis is concluded as a diagnosis and added to the history of the patient.
Therapy in the transplant population is clinically motivated or on a regular basis. The therapy expert executes the program based on the conclusion reached by the diagnostic expert. The therapy will base on the disease severity, the diagnoses reached and findings. The severity index is a measure of the degree of illness based on findings in the patient. The data for computing the severity index is obtained from surveillance visit, intensive care and etc. The therapy expert did not suggest a treatment plan for disease hypothesis which does not produce a positive test result.
Testing is done on the diagnostic and therapy expert systems. This is done by selecting certain medical cases from the archives of the University of Minnesota lung transplantation program. Before conducting the test, human expert is asked to highlight the data that were thought to be important in decision making. This include, positive/negative test, qualitative descriptors of patients (mild=1) and etc. The diagnostic expert correctly generated number of diseases parallel to the patient's medical record. The therapy expert also produces result in accordance with the surgical and drug recommendation that actually appear on the patient's medical record.
The diagnosis and therapy system successfully yielded accurate result based on the evaluation tests. The results demonstrate the feasibility of the proposed expert system. And this certainly will be useful for physician in decision making. There are some disadvantages in the system in which complex cases may result in complication of the system. Improvements in logic and knowledge representation are required to overcome the problem. Furthermore, evaluation on the system is tough due to the low number of real lung transplantation cases every year which is needed to conduct comparison between the system result and the medical record. This system may also cause the doctor to over rely on the Expert System. The result of the system can only be used as a reference, never to be used in making conclusion.
PRASAD, B. N. (11 June 1996). Artifical Intelligence in Medical Field. An Expert System For Diagnos And Therapy in Lung Transplantation.
Development of an optimized multi-biomarker panel for the detection of lung cancer based on principal component analysis and artificial neural network modeling
By: Nurul Nadiah Binti Adam
Lung cancer causes more deaths than any other cancer. It is crucial to detect lung cancer in early stage, where it is possible to reset the tumour and achieve healing. Computed tomography (CT) scan is commonly used to detect lung cancer. However, CT has a low specificity in sense that only a small percentage of nodule-positive patients will develop lung cancer, and furthermore, repeated radiation may promote carcinogenesis.
In recent years, artificial neural networks (ANN) have been suggested as auxiliary tools in medicine. ANN may play an important role in lung cancer by differentiating malignant from benign cells and to detect pulmonary nodules from CT chest images. ANNs are tools of artificial intelligence which intend to imitate the complex operation of organizing and processing information in the brain. It works by identifying patterns that correlate strongly a set of data which will correspond to a class by a learning process, in which interneuron connection weights are used to store knowledge of specific features identified within the data. Multilayered Perceptron (MLP) which is composed of three layers is a common ANN as shown in Figure 1:
Figure : Structure of Multilayer Perception (MLP) network used to classify lung cancer patients and controls with biomarker as input of the ANN
The information is entered from the input layer through the hidden and output layers of the network. The hidden and output layers are transformed by a validation function f(θ). Next, the value of the output is compared with a known target vector and the difference is computed as error. All of the set of biomarkers information was randomly divided as follows: 60% for training and 20% for validation. 20% of samples not used during the training of the ANN were used for testing. Principal component analysis (PCA) was applied to biomarkers that showed significant differences between groups in order to determine uncorrelated biomarkers that will better explain the variability observed in the data.
The advantage of this method is that biological markers that are pre-defined to be high-risk patients would enhance diagnostic capabilities and complete image studies since they are easily detectable in biological fluids using minimal invasive procedures. In addition, ANN had the best sensitivity at a specificity of 80%, compared with other nets and the best single marker. Besides that, ANN required fewer markers than needed by discriminant analysis in order to separate study groups, which directly reduced the cost involved. The disadvantage of this method is that when used alone, they show low sensitivity and specificity because lung cancer is a heterogenous disease.
In conclusion, the research paper successfully presented a strategy based on ANN technique to search for the best biomarker combination to distinguish lung cancer patients from control subject.
Prediction of postoperative morbidity after lung resection using an artificial neural network ensemble
By: Nur Hazirah Binti Mazlan
The objective of this study is to propose an ensemble model of artificial neural networks (ANNs) to predict cardiorespiratory morbidity after pulmonary resection for non-small cell lung cancer (NSCLC).
The proposed model for this study consists of a simple averaging artificial neural network ensemble [1,13,21,25]. A neural network ensemble is defined as a collection of a finite number of neural networks that is trained for the same target. In this study, the artificial neural network ensemble will merges the output of 100 different single artificial neural networks using simple average, so that the prediction of morbidity can be evaluated. They are two methods to implement ensembles which are Bagging and Boosting. These methods will give accurate results based on resampling techniques in the training phase. This simple averaging ensemble provides an effective scheme for predictions of the network.
Moreover, the 100 single artificial neural networks had slightly different structures and parameters. Each artificial neural network developed will be feed-forward multilayer perceptron networks with single hidden layer, trained with different backpropagation method. However, they will receive the same input and they only had one neuron in the output layer.
The backpropagation used for single artificial neural networks are; Levenberg-Marquardt, quasi-Newton, Powell-Baele conjugate gradient, Fletcher-Powell conjugate gradient, Polak-Ribiere conjugate gradient, resilient, and gradient descent with momentum and adaptive learning backpropagation. These training functions used with a gradient descent with momentum weight and bias learning function. The transfer functions used were hyperbolic tangent sigmoid functions and the performance function used was mean squared error performance function.
The ROC curve was constructed for the model by using the probability of compilation calculated by the artificial neural network ensemble.