Search Results

The Rapid MLS software device is designed to measure the midline shift of the brain from a NCCT acquisition and report the measurements. Rapid MLS analyzes adult cases using machine learning algorithms to identify locations and measurements of the expected brain midline and any shift which may have occurred. The Rapid MLS device provides the user with annotated images showing measurements. Its results are not intended to be used on a stand-alone basis for clinical decision-making or otherwise preclude clinical assessment of NCCT cases.

Rapid MLS software device is a radiological computer-assisted image processing software device using AI/ML. The Rapid MLS device is a non-contrast CT (NCCT) processing module which operates within the integrated Rapid Platform to provide a measurement of the brain midline. The Rapid MLS software analyzes input NCCT images that are provided in DICOM format and provides both a visual output containing a color overlay image displaying the difference between the expected and indicated brain midline at the Foramen of Monro; and a text file output (json format) containing the quantitative measurement.

Here's a breakdown of the acceptance criteria and the study that proves the device meets them, based on the provided FDA 510(k) Clearance Letter for Rapid MLS (K243378):

Acceptance Criteria and Device Performance

The core of the acceptance criteria for Rapid MLS appears to be its ability to measure midline shift with an accuracy comparable to or better than human experts.

Study Details

Here's a detailed summary of the study proving the device meets the acceptance criteria:

1. Sample Size Used for the Test Set and Data Provenance:

   • Sample Size: 153 NCCT cases
   • Data Provenance:
     • Country of Origin: Not explicitly stated for all cases, but sourced from 13 sites (2 OUS [Outside US], 11 US). This indicates a mix of international and domestic data.
     • Retrospective or Prospective: Not explicitly stated, but the description of "validation data was sourced and blinded independent of the development cases" and "demographic split for age and gender... used to test for broad demographic representation and avoidance of overlap bias with development" suggests these were pre-existing, retrospectively collected cases (i.e., not prospectively collected for this trial).
     • Scanner Manufacturers: Mixed from GE, Philips, Toshiba, and Siemens scanners.
     • Demographics: Male: 44%, Female: 56%, Age Range: 26-93 years.

2. Number of Experts Used to Establish the Ground Truth for the Test Set and Their Qualifications:

   • Number of Experts: 3 experts
   • Qualifications of Experts: Not explicitly stated, but the context implies they are medical professionals who use midline shift as a clinical metric, likely radiologists or neurologists.

3. Adjudication Method for the Test Set:

   • Method: Expert consensus was used to establish ground truth. The document states "ground truth established by 3 experts." This implies a consensus approach, but the specific method (e.g., majority vote, discussion to consensus) is not detailed. The "experts average pairwise MAE" suggests individual expert measurements were consolidated. It is not explicitly stated whether a 2+1 or 3+1 method was used, but given there were 3 experts, it's likely they reached a consensus view.

4. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study Was Done:

   • The study does compare the device's performance to human experts, but it's not explicitly described as a traditional MRMC comparative effectiveness study where human readers use the AI and then are compared to human readers without AI.
   • Effect Size of Human Readers Improvement with AI vs. Without AI Assistance: This specific comparison (human with AI vs. human without AI) was not the primary focus of the reported performance study. The study primarily evaluated the standalone performance of the AI in comparison to expert measurements (i.e., the AI as a "reader" vs. expert "readers"). The "Indications for Use" state that the results "are not intended to be used on a stand-alone basis for clinical decision-making or otherwise preclude clinical assessment of NCCT cases," implying it's an assistive tool, but the study described measures the AI's accuracy against experts, not the improvement of experts with the AI.

5. If a Standalone (Algorithm Only Without Human-in-the-Loop Performance) Was Done:

   • Yes. The document states, "Final device validation included standalone performance validation." The reported MAE of the Rapid MLS and its comparison to the experts' pairwise MAE directly reflect its standalone performance.

6. The Type of Ground Truth Used:

   • Ground Truth Type: Expert Consensus from the 3 experts.

7. The Sample Size for the Training Set:

   • Training Set Sample Size: 138 cases

8. How the Ground Truth for the Training Set Was Established:

   • The document implies that the "Algorithm development was performed using 162 cases from multiple sites; training included 24 cases for validation and 138 for training." While it doesn't explicitly state how ground truth was established for the training set, it is highly probable that a similar (if not identical) process involving human expert annotation was used, given the reliance on expert consensus for the validation/test set. The development cases were chosen to cover 0-18.6 mm offsets from expected midline, indicating a process of identifying and labeling the midline shift in these cases.

Ask a specific question about this device