CoLumbo

CoLumbo is an advanced software tool that utilizes a deep convolutional neural network to provide quantitative measurements and feature segmentation from lumbar spine MR images, aiding clinicians in their analysis and reporting workflow. The system emphasizes user control and responsibility, requiring clinicians to validate all software-generated outputs and apply their medical judgment, thereby functioning as a sophisticated measurement aid rather than an autonomous diagnostic system.

CoLumbo is a software medical device designed for radiologists and neuro- and spine surgeons to assist in the post-processing and analysis of lumbar spine Magnetic Resonance (MR) images. It processes previously-acquired DICOM images from adult patients aged 18 and above, excluding cases involving pregnancy, MRI scans with contrast media, post-operational complications, tumors, or infections. The software provides quantitative measurements of various spinal features by enabling feature segmentation, precise measurement, and threshold-based labeling of out-of-range values. Its primary output is an exportable report of these measurements, which users must review, potentially revise, and approve, intended to aid in visualizing, measuring, and documenting findings rather than for direct diagnosis or treatment recommendation. The core technology of CoLumbo utilizes a Deep Convolutional Image-to-Image Neural Network for its analytical capabilities, semi-automatically initiating adjustable measurements derived from anatomical segmentations. These segmentations serve as visual aids and a basis for calculation, not as final diagnostic outputs. Users retain full control and responsibility by defining individualized software settings including thresholds for identifying out-of-range measurements, confirming or modifying all software-generated segmentations and measurements, and ultimately interpreting the comprehensive data using their medical judgment and discretion. The software's performance has been validated through non-clinical testing, including a standalone assessment study involving 100 MR images compared against a ground truth established by three radiologists, and it incorporates cybersecurity measures to protect data integrity.

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. What specific deep learning architecture is employed in the "Deep Convolutional Image-to-Image Neural Network," and how was it trained and validated for diverse patient anatomies and pathologies not explicitly excluded?; How does the CoLumbo software handle variability in MR image acquisition parameters (e.g., sequence types, slice thickness, field strength) from different scanners, and what is its performance robustness across these variations?; The standalone performance assessment study compared CoLumbo outputs to a ground truth from 3 radiologists; what was the inter-rater reliability among these radiologists, and what were the specific performance metrics (e.g., Dice coefficient for segmentation, mean absolute error for measurements) achieved by the software?; How does the "rule-based algorithm" for modifiers interact with the neural network outputs, and to what extent can users customize these rules beyond simple threshold adjustments for identifying out-of-range measurements?; Given the exclusion of "post-operational complications, tumors, infections," what is the software's behavior or error handling when such conditions are inadvertently present in the input images, and are there any visual cues or warnings provided to the user in such scenarios?