Bartholmai, B J

Abstract Rationale In fungal endemic regions, differentiating benign from malignant pulmonary nodules on CT scans remains a major diagnostic challenge. Coccidioides nodules frequently resemble malignant lesions on CT and PET imaging, presenting significant challenges for differentiation, even among radiologists. Consequently, the benign biopsy and resection rate, especially in the Southwestern US, remains high. We previously developed [DOI:10.1371/journal.pone.0196910] and externally validated [DOI:10.1183/13993003.02485-2020] the Benign versus aggRessive nODule Evaluation using Radiomic Stratification (BRODERS) machine learning (ML) classifier to non-invasively distinguish between benign and malignant pulmonary nodules with an AUC of 0.87. However, its performance in certain benign nodules with malignant surface and texture characteristics like in coccidioidomycosis had a suboptimal AUC of 0.65. In this study, we redesigned our radiomics model by including numerous clinical features. Methods In this retrospective study, 113 pulmonary nodules (66 Coccidioides and 47 adenocarcinoma spectrum) detected on CTs with ground truth pathology diagnosis were identified by screening for key terms like “nodule”, “cocci”, and “adenocarcinoma” from both CT and pathology reports using a large language model trained on clinical textual data (Illuminate®). Clinical features such as demographics, smoking history, and laboratory studies were extracted by trained clinicians from the electronic medical record. All nodules were segmented and radiomic features extracted using a semi-automated seed growing algorithm previously described [DOI: 10.1097/JTO.0b013e3182843721]. The database was randomly split into an 80%-20% development and hold-out set prior to ML modeling. Feature selection for variables was conducted utilizing Elastic Net Regression, which provides a balanced approach between Lasso and Ridge regression. A logistic regression model was developed using the selected features and ground truth information of the training cohort. The true performance of the model was tested on the hold-out set. Results 160 radiomics and 22 clinical features were extracted. Elastic Net regression identified 44 features that contributed the most. Some of these clinical features include presence of satellite lesions, smoking history, and presence of positive IgG coccidiomycosis serology. The AUC of the ROC curve for the hold-out test set was 0.86. Conclusions The clinical features of significance that emerged during multimodal modeling are explainable. Designing explainable ML models is important, especially in the Southwestern US where Coccidioides nodules can appear to be malignant to both human experts and ML models. Although coccidioidomycosis serologies alone lack sensitivity and specificity to diagnose Coccidioides nodules, integrating them with a multimodal ML model can improve differentiation between benign Coccidioides and malignant adenocarcinoma nodules. This abstract is funded by: None

Abstract Rationale Interstitial lung disease (ILD) affects ∼50% of patients with systemic sclerosis (SSc) and is a leading cause of mortality. Visual assessment of ILD extent and morphologies, including ground-glass opacities (GGO) and reticular densities (RD), suffers from poor interobserver agreement, limiting prognostic consistency. CALIPER Quantitative CT (QCT) provides reproducible volumetric measures of parenchymal and vascular features. This study evaluated baseline QCT-derived lung features for association with transplant/mortality risk in SSc-ILD. Methods A retrospective cohort of SSc patients (≥18 years) seen at an academic center from 2007-2018 was identified using ICD codes, with ACR/EULAR 2013 SSc criteria confirmed by chart review. Inclusion required ILD on high-resolution CT (HRCT) ±2 years of SSc diagnosis, and PFTs ±6 months of HRCT. CALIPER quantified volumes of normal lung, GGO, RD, hyperlucency, honeycombing (HC), and pulmonary vascular-related structures (PVRS), which are expressed as percentages of total lung volume. Right heart catheterization confirmed pulmonary hypertension (PH) defined by 2022 ESC/ERS criteria (Eur Heart J, 2022). Cox hazards models assessed associations with combined mortality/transplant outcome, a priori adjusted for age, sex, diffuse SSc subtype, ILD duration, and PH (Column B), and further adjusted for baseline percent predicted FVC (ppFVC) and ppDLCO (Column C)(Table 1). Results Of 132 SSc-ILD patients, median SSc and ILD disease durations at HRCT were 1.4 (0.6, 2.4) and 0.0 (0.0, 0.5) years. Median feature percentages: total ILD 13.0% (IQR: 6.4, 21.0), GGO 10.2% (4.6, 16.9), RD 1.9% (0.8, 3.7), HC 0.03% (0.00, 0.09), PVRS 3.2% (2.3, 4.1). Median ppFVC was 81.1% (69.4, 92.6), and ppDLCO 60.0% (46.0, 75.7); 21 patients had PH. There were 2 transplants and 45 deaths. Every 5% increase in baseline total ILD and GGO associated with 22% (HR 1.22; 95% CI 1.12-1.33) and 24% (1.24; 1.12-1.36) higher risk of transplant/mortality on unadjusted models (Column A), respectively, remaining significant after a priori (CB) and full adjustment (CC)(Table 1). Every 1% increase in RD and PVRS conferred 19% (1.19; 1.08-1.31) and 33% (1.33; 1.08-1.63) higher risk of transplant/mortality in unadjusted models (CA); RD retained significance after a priori adjustment (1.17; 1.03-1.33), while PVRS trended non-significant in adjusted models. Conclusions In early SSc-ILD, QCT-derived GGO, RD, and PVRS provide robust prognostic value for mortality/transplant risk assessment, with small increases in baseline severity—at or below visual detection thresholds—associating with clinically meaningful risk increases. These findings support integrating baseline QCT biomarkers into risk stratification beyond clinical and PFT measures. This abstract is funded by: Rheumatology Research Foundation