CoLumbo

CoLumbo is an image post-processing software tool that employs a Deep Convolutional Image-to-Image Neural Network to derive quantitative measurements and segment anatomical features from lumbar spine MRIs, aiding clinicians in their review and analysis. The device aims to improve the efficiency of visualizing, measuring, and documenting spinal features, while the user maintains control over settings, verification of outputs, and ultimate diagnostic interpretation.

CoLumbo is a medical image post-processing and measurement software tool designed to assist radiologists, neurosurgeons, and spine surgeons in analyzing lumbar spine Magnetic Resonance (MR) images. Its core function is to provide quantitative measurements from previously acquired DICOM images for users to review, analyze, and interpret. The software facilitates this through feature segmentation, feature measurement, threshold-based labeling of out-of-range measurements (based on user-defined settings), and the export of these measurement results into a written report for user review, revision, and approval. CoLumbo is intended for use in hospitals and medical institutions on DICOM MRI images from lumbar spine exams of patients aged 18 and older, excluding those who are pregnant, underwent MRI with contrast media, or have post-operational complications, scoliosis, tumors, infections, or fractures. The software explicitly does not provide diagnoses or treatment recommendations, emphasizing that the user retains full responsibility for medical judgment. The underlying technology of CoLumbo involves a Deep Convolutional Image-to-Image Neural Network for its semi-automatic segmentation capabilities. A standalone software performance assessment study was conducted in the U.S. using 101 MR image studies from diverse patient demographics and MRI systems from five manufacturers. This study compared CoLumbo's outputs to a ground truth established by three U.S. radiologists. The primary endpoint focused on measurement accuracy, demonstrating Mean Absolute Errors (MAE) within predefined limits for key metrics such as Dural Sac Area (14.8 mm²), Lordotic Angle (2.6°), and Listhesis/AP Slip (0.9 mm). Secondary endpoints for other measurements and segmentation accuracy (Mean Dice Coefficient, MDC) were also met, with MDC values like 0.97 for Disc/Vertebral Body (Axial) and 0.92 for Dural Sac (Axial). The study confirmed that this accuracy was maintained across critical subgroups, including MRI scanner manufacturer, race, sex, and patient age, and that the training, testing, and validation datasets for the machine learning algorithm were independent.

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. How does the performance of CoLumbo's Deep Convolutional Image-to-Image Neural Network compare to other AI models used for similar lumbar spine image analysis tasks in terms of segmentation accuracy (MDC) and measurement precision (MAE)?; What is the variability in CoLumbo's quantitative measurements when applied to images from different MRI scanner field strengths (e.g., 1.5T vs. 3.0T) beyond the specific manufacturers tested, and how might this impact clinical interpretation?; Given the exclusion criteria (e.g., post-operational complications, scoliosis, tumors), what is the software's behavior or failure mode if such images are inadvertently processed, and are there safeguards?; How sensitive are the segmentation and measurement outcomes to variations in MRI acquisition protocols (e.g., slice thickness, sequence parameters) commonly encountered in clinical practice across different institutions?; What is the learning curve for radiologists and surgeons to effectively utilize CoLumbo, particularly in setting user-defined thresholds and verifying/editing software-generated segmentations and measurements?