Quality Assurance Review by Pro Surgical 3D Segmentation:See How the AMOS22 Grand Challenge Data Performs
We applied the quality assurance toolset provided within ProSurgical3D Segmentation to find errors in the AMOS22 Grand Challenge training dataset. Take a moment to see ProSurgical3D Segmentation in action and quickly reveal problematic scans, segmentations, and labeling errors. Our segmentation tool offers a seamless segmentation experience, combining user-friendly tools with advanced quality assurance capabilities. Enhance your data quality, optimize your team's productivity, and improve your machine learning predictions by eliminating errors in your training data.
Scans 517, 518, and 540 have significant problems and should be removed from the training set.
LIST OF ERRORS
Scan 49 Stomach is over segmented and appears to include additional non-stomach regions.
What did we find?
Scan 317 Left Adrenal Gland has a poor segmentation issue.
We used the Segmentation for Pro Surgical 3D for 30 minutes and found all the issues shown to the right. Take a look at the videos below to see how we did it.
Scan 592 Aorta has a significant segmentation error
The MRI nii files do not indicate the affine transformation to flip align the X axis
And more. Watch the videos below to see what we found.
AI & ML-Powered 3D Image Annotation: Streamline Your Workflow
Revolutionary 3D Image Segmentation: Next-Gen Speed, Precision, and Ease
1. Importing AMOS 22 Data
Prepare your dataset for quality assurance analysis by seamlessly importing your pre-existing segmentation and labeled data.
2. Optimizing Data Integrity
After completing the import process, review some scans to confirm accuracy. Create an attribute label database for organ types and left/right sides. Assign these attributes to scan objects for consistency across your dataset.
3. Evaluating Training Sets/Quality Assurance
After assigning attributes, use the quality assurance tools to analyzing the objects, looking for outliers, and ranking them based on various metrics. Identify problematic scans and objects, such as unusual volume and segmentation errors, and remove or correct them for better data quality.
4. Investigating Anomalies: Mirrored Objects and MRI-Like Scans in CT Data
Examine outliers to understand data issues. See when objects are mirrored and identify mixed modality issues when the scan set includes both MRI and X-ray CT scans. This anomaly becomes evident when reviewing the scan sets and outliers.
5. Enhancing Data Quality: Evaluating CT Scan Outliers with Corrected Attributes
After assigning correct attributes to the MRI scans, we return to the quality assurance review. We filter out MRI scans, focusing on CT scans. We review outliers based on attributes like organ and side and explore them individually for potential segmentation issues or anomalies, allowing for better data quality and expert review if needed.
6. Detecting Outliers in CT Scans for Disease Indicators
When reviewing CT scans, it's important to identify outliers in different object classes by examining values significantly deviating from the norm. For instance, the left kidney shown in this video stands out as a potential anomaly, possibly indicating disease. To address this, consider adding disease-related attributes or conduct specialized analysis.
7. Exported Labeled Data
Export your improved quality data labels and its associated meta data. Labeled data is exported into common file formats so that you can feed these into your machine learning training pipelines. Learn how to access crucial details like file indexing, spatial properties, and object attributes for enhanced insights. See how this information can supercharge your machine learning endeavors and elevate your data analysis game!
8. Feed into your ML Training Pipeline
Discover how to effectively use the scan CSV file to index your scan files and gain access to vital details such as file origin, spatial properties, and object attributes. This video is a must-watch for anyone seeking to harness the power of data analysis and improve their machine learning processes
High Quality Data
Better data quality, better outcomes
Experience streamlined 3D segmentation and labeling with our expert-approved machine learning algorithms, saving time and resources. Our 3D segmentation tool offers a simple and user-friendly interface, with visual feedback and advanced features, and robust validation tools making it easy for anyone to process 3D image segmentations and labels accurately and efficiently.
