Comparison of radiologist observations prior to preprocessing.

Radiologist observations prior to preprocessing for machine learning were compared by treatment outcome. P-values were calculated for continuous variables using analysis of variance test. P-values for categorical variables were calculated using Chi-squared test. The following variables were dropped from further analysis: Anomalymediastinumvesselsdevelop, shadowpattern, affectlevel, thromboembolismpulmonaryartery, anomalylungdevelop, and accumulationcontrast. The following variables were refactored (S3 Table) to recombine levels: Lungcavitysize, affectlevel, totalcavernum. Glossary: Affectlevel–location of affected lung area; affectpleura—changes in the pleura; bodysite_coding_cd–which lung is the observation located; bronchialobstruction—bronchial obstruction syndrome disorders, dissemination—Diffuse pulmonary nodules detected; limfoadenopatia–greater than 10 mm is considered the upper limit for normal nodes (short transverse diameter); lungcapacitydecrease—reduced lung volumes; lungcavitysize–size of lung cavity; nodalcalcinatum—Nodi Calcinatum detected; plevritis—pleural effusion detected; pneumothorax—Pneumothorax detected; posttbresiduals—Post-tuberculosis changes in the lung; processprevalence–prevalence of process in number of segments; totalcavernum–number of cavities; thromboembolismpulmonaryartery—Thromboembolism Of The Pulmonary Artery detected; anomalymediastinumvesselsdevelop—Anomaly Of Mediastinum Vessels Develop detected; shadowpattern–shadowpattern of nodule, node, or infiltrate; affectedsegments–segments of lung that are affected; accumulationcontrast–amount of contrast accumulated.
(XLSX)