| 1 Definition | 2 Functional Requirements | 3 Syntax | 4 Semantics |
1 Definition
The Medical Image Processing Type defines the allowable operations, methods, and processing descriptors used for Medical Imaging data (e.g., MRI, CT, PET, Ultrasound).
It reuses Common Definitions for:
- Header
- Algorithm
- Algorithms
- FeatureClass
- Features.
2 Functional Requirements
The Medical Image Processing Type shall:
- Fix Domain = MedicalImaging.
- Validate Operation against medical‑imaging‑specific enumerations.
- Validate Method against recognised classical, ML‑based, and deep‑learning imaging techniques.
- Allow Algorithm to be a string ID or an AlgorithmObject.
- Allow Algorithms to be an array of Algorithm items.
- Require Features to be a non‑empty array of unique strings.
3 Syntax
https://schemas.mpai.community/AIH1/V1.0/data/MedicalImageProcessingType.json
4 Semantics
| Label | Description |
|---|---|
| Header | Medical Image Processing Type Header, Standard “AIH‑MPT‑Vx.y”. |
| Domain | Constant value "MedicalImaging". Applies exclusively to medical images. No other value permitted. |
| Operation | Specifies the medical‑image processing step. Enumerated list includes: Segmentation, Registration, Denoising, Enhancement, LesionDetection, Classification, Quantification, Reconstruction, MotionCorrection, FeatureExtraction. |
| Segmentation | Partitioning an image into anatomical or functional regions (e.g., organs, lesions). |
| Registration | Aligning one or more images into a shared spatial reference frame. |
| Denoising | Reducing noise while preserving clinically relevant structures. |
| Enhancement | Improving contrast, sharpness, or visibility of image content. |
| LesionDetection | Identifying tumors, nodules, or abnormalities. |
| Classification | Assigning diagnostic or morphological labels to images or ROIs. |
| Quantification | Computing measurable biomarkers (volume, density, radiomics statistics). |
| Reconstruction | Producing images from raw sensor data (e.g., MRI k‑space, CT sinograms). |
| MotionCorrection | Correcting motion artefacts from patient or physiological movement. |
| FeatureExtraction | Deriving radiomic, deep‑feature, or structural descriptors from images. |
| Method | Processing technique used to implement the operation. Must be one of: OtsuThresholding, UNet, ResNet, FLIRT, ANTs, SIFT, SURF, NonLocalMeans, BilateralFilter, HistogramEqualization, CNN, Transformer, LevelSet, GraphCut. |
| OtsuThresholding | Global thresholding based on histogram variance minimisation. |
| UNet | CNN model for segmentation of medical images. |
| ResNet | Deep residual network for classification or feature learning. |
| FLIRT | Linear registration tool (FSL). |
| ANTs | Nonlinear registration and normalization framework. |
| SIFT | Local keypoint detector and descriptor for matching. |
| SURF | Fast keypoint detector/descriptor for feature matching. |
| NonLocalMeans | Patch‑based denoising preserving anatomical detail. |
| BilateralFilter | Edge‑preserving smoothing filter. |
| HistogramEqualization | Intensity redistribution to enhance contrast. |
| CNN | Convolutional neural network‑based image processing. |
| Transformer | Attention‑based deep‑learning model for medical imaging. |
| LevelSet | Contour‑evolution method for segmentation. |
| GraphCut | Graph‑optimization‑based segmentation technique. |
| Algorithm | String identifier or an AlgorithmObject defined in CommonDefinitions. |
| AlgorithmObject.Name | Required name of the algorithm (e.g., “UNet‑Baseline”, “ANTs‑SyN”). |
| AlgorithmObject.Version | Optional version ID. |
| AlgorithmObject.Params | Free‑form configuration parameters for the algorithm. |
| Algorithms | Array of Algorithm entries (string ID or AlgorithmObject). |
| FeatureClass | Category describing extracted imaging features (e.g., radiomics, morphologic, deep features). |
| Features | Non‑empty array of unique feature names (e.g., lesion_volume, entropy, Hounsfield_mean). |
| Trace | Provenance information and Time. |
| DescrMetadata | Descriptive Metadata |