Key points are not available for this paper at this time.
Background: Differentiating infective from malignant pulmonary lesions on routine chest CT often requires invasive procedures and may suffer from interobserver variability.This study compares expert radiological assessment with a quantitative radiomic pipeline to determine their relative diagnostic performance.Materials and methods: 30 patients (15 infective, 15 malignant) with histopathological or microbiological confirmation underwent chest CT.Two trained thoracic radiologists independently reviewed all scans, classifying each lesion as benign or malignant.Separately, semiautomatic 3D Slicer segmentation was performed, and 85 IBSI-compliant radiomic features (shape, first-order, GLCM, GLDM, GLRLM, GLSZM) were extracted via PyRadiomics.Performance metrics sensitivity, specificity, accuracy, and area under the ROC curve (AUC) were calculated for both approaches, with significance assessed by Fisher's exact test (p < 0.001).Results: Radiologists correctly identified all 15 benign lesions (specificity = 100%) and 14 of 15 malignant lesions (sensitivity = 93.3%),yielding 96.7% accuracy and AUC = 0.967.The radiomic model achieved perfect sensitivity (100%), correctly detecting all malignancies, and 93.3% specificity (14/15 benign), with 96.7% accuracy and an AUC of 0.967.Both methods' associations with true lesion status were highly significant (p ≈ 2 × 10 -7 ).Conclusion: Expert radiology and radiomic analysis demonstrate equivalent overall accuracy and discrimination (AUC ≈ 0.97) but exhibit complementary strengths.Radiologists maximize specificity, whereas radiomics ensures no malignancies are missed.A hybrid workflow leveraging radiomic screening followed by radiologist review may optimize diagnostic safety and efficiency.
Mathi et al. (Mon,) studied this question.