Diagnostic accuracy of computer-aided detection (CAD)-derived solid component proportion and mean CT attenuation for...

Accurate preoperative identification of invasiveness in pulmonary part-solid nodules (PSNs) is critical to guide extent of resection and operative timing, yet reliable, automated imaging markers are limited. This study aimed to evaluate the predictive value of volumetric proportion and mean computed tomography (CT) value of solid components for the invasiveness of pulmonary PSNs using computer-aided detection (CAD).

This retrospective study included 313 patients with 324 PSNs who underwent thin-section CT (TSCT) before surgery and had postoperative pathologic diagnoses of atypical adenomatous hyperplasia (AAH), adenocarcinoma in situ (AIS), minimally invasive adenocarcinoma (MIA), or invasive adenocarcinoma (IA). The proportion of solid components at CT values of -300 Hounsfield units (HU) and -400 HU and the mean CT values of the solid components were automatically computed. The diagnostic accuracy was evaluated with the area under the receiver operating characteristic curve (AUC). The tumor size was stratified, and the optimal predictive indicators within each stratum were evaluated.

Among 313 patients (324 PSNs; median age 58 years), 4 AAH, 88 AIS, 69 MIA, and 163 IA were identified. The mean CT value of the solid component had an AUC of 0.795 (threshold -144.5 HU), with sensitivity 0.799, and specificity 0.696. Tumor size had an AUC of 0.784 (threshold 1.65 cm, sensitivity 0.595, specificity 0.826), and solid component size had an AUC of 0.759 (threshold 0.95 cm, sensitivity 0.804, specificity 0.509). Analysis of solid component proportion demonstrated AUCs of 0.795 and 0.783 at -300 and -400 HU, respectively, with corresponding optimal thresholds of 16.2% (sensitivity 0.706, specificity 0.814) and 31.3% (sensitivity 0.687, specificity 0.807). In the stratified analysis, the mean CT value and solid component proportion showed high AUCs in all tumor size strata.

CAD-derived volumetric solid proportion and mean CT attenuation show promise for predicting PSN invasiveness in this single-center retrospective cohort; prospective multicenter validation with standardized protocols is warranted.