Abstract Rationale Photon-counting computed tomography (PCCT) is an emerging imaging technology that advances traditional high-resolution CT (HRCT) by detecting and counting individual X-ray photons and measuring their energy levels. This enables higher spatial resolution, lower radiation dose, reduced image noise, and improved material decomposition. To maintain continuity and comparability between conventional and advanced CT imaging in interstitial lung disease (ILD), our goal was to evaluate the performance of existing quantitative techniques based on machine learning (ML) and deep learning (DL) algorithms when applied to PCCT data. Methods Twenty-seven subjects who underwent PCCT using a diffuse lung disease protocol were retrospectively collected. Previously developed ML and DL algorithms—based on radiomic features and a residual convolutional neural network (CNN), respectively—were used to compute quantitative ILD scores. The quantitative lung fibrosis (QLF) and quantitative ILD (QILD) scores represented fibrotic and total ILD burden, respectively. A thoracic radiologist verified the ML-derived quantitative scores after reviewing the corresponding PCCT scans. To compare quantitative results between ML and DL models, paired equivalent t-tests were conducted. Results All 27 PCCT scans were anonymized and processed through the end-to-end ML and DL models. The mean (±SE) difference between ML and DL QLF scores was -1.01% (±0.29). The ML and DL QLF scores demonstrated equivalence within a 2% limit (p = 0.001). This finding indicates that both ML- and DL-based quantification methods yield comparable results in assessing ILD extent on PCCT. Conclusions Quantitative ILD scores derived from PCCT can be reliably obtained using both ML and DL algorithms. The results demonstrate high consistency between the two methods in estimating fibrotic and total ILD burden. DL-based quantitative scores appear robust for volumetric HRCT or PCCT scans and show strong agreement with radiomic ML-derived metrics. Implementation of DL quantitative analysis in cloud-based or web-deployed environments can facilitate rapid, automated ILD assessment, enabling efficient patient screening, cohort enrichment in clinical trials, and advancement of precision medicine applications using real-world imaging data. This abstract is funded by: Siemens
Kim et al. (Fri,) studied this question.