Optellum Virtual Nodule Clinic, Optellum software, Optellum platform

The Optellum™ Virtual Nodule Clinic is a software platform employing a deep learning-based computer-aided diagnosis (CADx) tool, the LCP-CNN, to assist clinicians in characterizing incidentally detected pulmonary nodules from CT scans by providing a risk score. This system aims to improve diagnostic accuracy and support clinical decision-making in nodule management by analyzing morphological features within the nodules using an artificial intelligence algorithm trained on a dataset of known benign and malignant cases.

The Optellum™ Virtual Nodule Clinic (VNC) is a software device designed for tracking, assessing, and characterizing incidentally detected pulmonary nodules. It features a computer-aided diagnosis (CADx) function, exclusively for pulmonologists and radiologists, which utilizes an artificial intelligence algorithm called the Lung Cancer Prediction Convolutional Neural Network (LCP-CNN). This algorithm analyzes user-selected regions of interest (ROI) within lung CT data, extracting imaging features based on morphological characteristics to compute a single integer LCP-CNN score from 1 (very likely benign) to 10 (very likely malignant). This score is presented with a histogram comparing it to scores from a database of cases with known ground-truth, intended as one input for clinical decision-making. Other users, like clinicians and nurses, can use VNC for patient management workflow, tracking, and viewing images/reports, but without access to the LCP-CNN score. The LCP-CNN score is indicated for evaluating incidentally detected solid and semi-solid pulmonary nodules (5-30mm diameter) in patients aged 35 years or older; multiple nodules can be assessed independently. Contraindications include use for lung cancer screening, pure ground glass opacities, calcified nodules, patients with a cancer history within the last 5 years, more than five nodules, or thoracic implants affecting image appearance. A clinical study involving 12 readers (radiologists and pulmonologists) and 300 CT cases demonstrated that using the LCP-CNN score improved the average diagnostic accuracy of readers in discriminating between malignant and benign nodules by 6.85 AUC points (from an average unaided AUC of 81.9 to an aided AUC of 88.8). The VNC software also integrates with hospital PACS and RIS for read-only access to images and reports.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. The LCP-CNN training data comprised 8% malignant nodules and the mapping dataset for the 1-10 score used 10% malignant cases, while the validation dataset presented to readers had a 50% malignancy rate. How does the system account for, or guide users to adjust their interpretation of the LCP-CNN score in real-world clinical scenarios where the prevalence of malignancy in incidentally detected nodules might be significantly lower (e.g., <10%) than in the validation set?; What specific morphological or radiomic features (e.g., spiculation, attenuation, texture heterogeneity) are most heavily weighted by the LCP-CNN algorithm in generating its score, and is there any mechanism for clinicians to understand the image-based drivers for a particularly high or low score on a case-by-case basis?; The MRMC study demonstrated an average AUC improvement of 6.85 points. How did this translate into specific changes in management recommendations by the readers, such as the proportion of cases where follow-up was appropriately de-escalated for benign nodules or escalated for malignant nodules, and what was the net reclassification improvement at clinically relevant risk thresholds?; Given the exclusion of high contrast CT images (>300HU in aortic arch), images with implants, motion artifacts, or missing slices from clinical validation, how does the LCP-CNN algorithm's performance specifically degrade or is its use actively restricted when encountering such images, which may still be presented to the user if not explicitly contraindicated for the LCP-CNN score?; The device is indicated for nodules 5-30mm. How does the LCP-CNN score performance vary across this size range, particularly for smaller nodules (e.g., 5-8mm) where inter-reader variability and diagnostic uncertainty are often highest, and were there significant differences in AUC improvement for these smaller nodules in the MRMC study?