China
Africa
Explore chapters and articles related to this topic
Identifying Adverse Events
Published in Mohamed Elgendi, PPG Signal Analysis, 2020
Unlike sensitivity and specificity, positive/negative predictive values (and their derived metrics) are dependent on the prevalence of the condition in a population. Here is a MATLAB function that can help with evaluating an application’s performance:
Machine Learning
Published in Pedro Larrañaga, David Atienza, Javier Diaz-Rozo, Alberto Ogbechie, Carlos Puerto-Santana, Concha Bielza, Industrial Applications of Machine Learning, 2019
Pedro Larrañaga, David Atienza, Javier Diaz-Rozo, Alberto Ogbechie, Carlos Puerto-Santana, Concha Bielza
Classification accuracy measures the fraction of instances correctly classified by the classification model. Conversely, error rate measures the proportion of misclassifications. Thus, Acc(ϕ)+Err(ϕ)=1 $ \mathrm Acc (\phi ) + \mathrm Err (\phi ) = 1 $ . Sensitivity, also known as recall or the true positive rate (TPR), represents the proportion of true positives successfully detected by the classifier. Specificity is defined similarly for true negatives. The false positive rate (FPR) is one minus specificity. The positive predictive value, also known as precision, measures the proportion of correctly assigned positive instances. The negative predictive value is defined similarly for negative instances. The F1 $ \boldsymbol{F}_1 $ measure is the harmonic mean of the precision and recall measures. Cohen’s kappa statistic (Cohen, 1960) first corrects the accuracy measure considering the result of a chance match between the classifier, ϕ(x) $ \phi (\mathbf{{x}}) $ , and the label generation process, C. The bottom row of Table 4 shows the numerator, where the expected proportion of matched instances under the null hypothesis of independence between the true class and the predicted class (mere chance) is subtracted from the classification accuracy. Then the measure is normalized between 0 and 1, as specified in its denominator. All eight performance measures above take values in the interval [0, 1]. Values close to 1 are preferred for all the measures, except for error rate. Values close to 0 are better for error rate.
A novel method for detection of COVID-19 cases using deep residual neural network
Published in Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 2021
Ali Noshad, Parham Arjomand, Ahmadreza Khonaksar, Pooya Iranpour
For the purpose of validation, chest X-ray scans of 19 patients including 13 cases with Polymerase Chain Reaction (PCR) confirmed diagnosis of COIVD-19 pneumonia, pulmonary oedema, and normal cases obtained from Namazi hospital in Iran were retrospectively reviewed by an experienced radiologist and DRCOVID-Net blinded to the final diagnosis. Radiologist accurately diagnosed 9/13 cases (reported either as definitely or possibly positive cases) with an accuracy of 79%, while the proposed model able to correctly identify 12/13 COVID-19 cases with an accuracy of 84%. Moreover, the proposed model was able to achieve a sensitivity of 92.31%, whereas, an expert radiologist was reached to a 69.23% mark. A highly sensitive model demonstrates positive results in the individuals infected with the virus, yet, positive diagnosis can occur in the negative cases. However, it would hardly ever be negative in a person with a disease. Hence, if a highly sensitive test is negative, it almost unmistakably rules out the disease. In our cases, positive and negative predictive values are directly related to the prevalence of the disease. Considering all other factors remain constant, the Positive Predictive Value (PPV) will rise with increasing prevalence; and Negative Predictive Value (NPV) reduces with an increase in prevalence. Predictive values refer to the ability of the test results to confirm the presence or absence of a disease, based on whether it is positive or negative, respectively. PPV reflects the probability that an individual with a positive test result truly has the disease, which the proposed model achieved the value of 85.71%. NPV is the probability that an individual with a negative test result truly does not have the disease, and in our case, the model was outperformed the radiologist and reached to a value of 80.00%. Table 2 shows the statistical analysis on the obtained results from an expert radiologist and the proposed model.