Utility
- Patients can use this tool to check risk of pneumonia with only seven blood tests. It does not require any radiological examination.
- Doctors can also use it for pneumonia risk prediction. It can enable them to order further investigations for patients with high risk to provide improved treatment.
- It can also be used by COVID-recovered patients to assess their post-COVID lung health regularly.
Dataset
The dataset contains 247 COVID-19 patients from the publicly available dataset published by Feng, Z. et al. (2020) which was originally used to predict the risk of progression of coronavirus pneumonia. The acquired data was split randomly into training (n = 222) and validation (n = 25) datasets in 90:10 ratio for training and validating the regression models respectively. To check out the datasets, please go to the Datasets page.
Features used :
- Age
- Neutrophil-lymphocyte ratio (NLR)
- Aspartate aminotransferase (AST)
- Albumin
- Lactic dehydrogenase (LDH)
- C-reactive protein (CRP)
CT severity score
Feng, Z. et al. (2020) proposed a method to compute CT (computed tomography) severity score that evaluates the extent of lesions in every lung lobe. Each lobe is scored in the range 0 to 5 as per the table given below. The score of each of the five lung lobes is added to get the CT severity score in the range 0 to 25.
| Score | Criteria |
| 0 | No involvement of lesions |
| 1 | ≤ 5% involvement |
| 2 | 6-25% involvement |
| 3 | 26-50% involvement |
| 4 | 51-75% involvement |
| 5 | >75% involvement |
Prediction methodology
Support Vector Regression (SVR) and Multi-layer Perceptron Regression (MLPR) models were trained using the patient data in the training dataset to compute the CT severity score. Training was performed using 5-fold cross validation. The models were externally validated using the data in validation dataset. The density of CT severity scores was computed for patients with high risk (fhigh) and low risk (flow) of pneumonia. Confidence of low risk of pneumonia was computed as :
and confidence of high risk of pneumonia as :
where s is the predicted CTSS. The confidence values of risk of pneumonia for different predicted CT severity scores are given in the table below :
| Predicted CT severity score | Confidence range of low risk of pneumonia (%) | Confidence range of high risk of pneumonia (%) | Absolute difference between high and low risk of pneumonia (%) |
| 0-2 | 99.39 - 90.64 | 0 - 1.64 | 99.39 - 89 |
| 2-4 | 88.57 - 71.28 | 2.02 - 5.08 | 86.55 - 66.2 |
| 4-6 | 68.69 - 51.62 | 5.54 - 8.98 | 63.15 - 42.64 |
| 6-8 | 49.27 - 33.36 | 9.61 - 15.91 | 39.66 - 17.45 |
| 8-10 | 31.24 - 19 | 17.14 - 27.91 | 14.1 - 8.91 |
| 10-12 | 19 - 10.95 | 27.91 - 40.37 | 8.91 - 29.42 |
| 12-14 | 10.95 - 3.99 | 40.37 - 51.3 | 29.42 - 47.31 |
| 14-16 | 3.99 - 1.26 | 51.3 - 65.89 | 47.31 - 64.63 |
| 16-18 | 1.26 - 0.28 | 65.89 - 78.77 | 64.63 - 78.49 |
| 18-20 | 0.28 - 0 | 78.77 - 85.17 | 78.49 - 85.17 |
| 20-22 | 0 - 0 | 85.17 - 89.27 | 85.17 - 89.27 |
| 22-25 | 0 - 0 | 89.27 - 96.97 | 89.27 - 96.97 |
For predicted CT severity scores of less than 6, confidence of low risk of pneumonia is greater than 50%. For predicted CT severity scores of greater than 14, confidence of high risk of pneumonia is greater than 50%. For scores between 6 and 14, confidences of both high and low risk of pneumonia are below 50%. Thus, the classification can be made by selecting the maximum confidence for such scores.
Bhattacharjee S. et al., Biomed Signal Process Control (2022), 77:103745. doi: 10.1016/j.bspc.2022.103745. Please contact Dr. Sudipto Saha (ssaha4@jcbose.ac.in) regarding any further queries.