| name | matlab-image-processing-toolbox-v2 |
| description | MATLAB Image Processing Toolbox. Functions - imgaussfilt, medfilt2, wiener2, imfilter, graythresh, imbinarize, multithresh, watershed, activecontour, strel, imopen, imclose, imerode, imdilate, bwareaopen, imfill, regionprops, bwconncomp, bwlabel, edge, im2double, im2uint8, mat2gray, adapthisteq, imadjust, blockproc. Tasks - remove noise from an image, filter a noisy image, smooth an image, enhance contrast, threshold an image, segment objects, separate touching objects, clean up a binary mask, fill holes in mask, remove small objects, count cells or particles, measure region properties like area and centroid, detect edges, convert image data types, preprocess images before deep learning, apply morphological operations, extract texture features, process large images in blocks. Domains - MRI preprocessing, CT windowing, microscopy, histology, cell counting, fluorescence imaging, image segmentation, satellite imagery, industrial inspection. |
MATLAB Image Processing Toolbox
Expert skill for medical image analysis using MATLAB's Image Processing Toolbox (IPT R2025a+). Focuses on medical imaging patterns, domain-specific pipelines, and non-obvious gotchas that go beyond standard IPT knowledge.
When to Use This Skill
- Medical imaging pipelines: MRI, CT/X-ray, microscopy, histology
- Modality-specific denoising (Rician for MRI, speckle for ultrasound)
- Multi-tissue segmentation with domain-specific thresholds
- Cell counting, fluorescence quantification, stain normalization
- DL segmentation post-processing with IPT morphological cleanup
- Data type gotchas causing silent errors in medical image processing
Read Before Coding
| Task | Read This | Template Script |
|---|
| MRI preprocessing | knowledge/cards/medical-mri.md | scripts/template_mri_preprocessing.m |
| Cell counting | knowledge/cards/medical-microscopy.md | scripts/template_cell_counting.m |
| CT windowing | knowledge/cards/medical-mri.md | scripts/template_ct_windowing.m |
| Fluorescence quantification | knowledge/cards/medical-microscopy.md | scripts/template_fluorescence_quantification.m |
| Histology stain normalization | knowledge/cards/medical-microscopy.md | scripts/template_histology_stain_normalization.m |
| Tissue thresholding | knowledge/cards/segmentation-thresholding.md | scripts/template_adaptive_thresholding.m |
| Watershed for touching cells | knowledge/cards/medical-microscopy.md | scripts/template_watershed_segmentation.m |
| Morphological mask cleanup | knowledge/cards/morphology-binary.md | scripts/template_morphological_cleanup.m |
| DL segmentation + IPT | knowledge/cards/deep-learning-segmentation.md | โ |
| Data type gotchas | knowledge/cards/data-types.md | โ |
| Modality-specific denoising | knowledge/cards/filtering-denoising.md | โ |
| Advanced regionprops | knowledge/cards/feature-regions.md | โ |
| Large image processing | โ | scripts/template_large_image_blockproc.m |
| Edge detection pipeline | โ | scripts/template_edge_detection_pipeline.m |
Critical Rules
Rule 1: imfilter Zero-Pads by Default (Creates Dark Borders)
% imfilter defaults to zero-padding โ dark border artifacts!
bad = imfilter(I, h); % Dark borders
good = imfilter(I, h, 'replicate'); % Fix
% imgaussfilt defaults to 'replicate' โ safe as-is
filtered = imgaussfilt(I, 2); % Already correct
Rule 2: graythresh Returns [0,1], Not Pixel Values
level = graythresh(img_uint8); % Returns 0.45, NOT 115
% Use imbinarize instead (handles automatically)
bw = imbinarize(img_uint8);
Rule 3: double() vs im2double() โ Silent Disaster
bad = double(img_uint8); % Still [0,255], NOT [0,1]!
good = im2double(img_uint8); % Correctly scales to [0,1]
Rule 4: SE Radius Must Be Smaller Than Features You Keep
% SE radius โ half the feature size you want to AFFECT
se = strel('disk', 5); % Removes features ~10px diameter
cleaned = imopen(bw, se);
% If SE radius > feature radius, features disappear!
Rule 5: blockproc BorderSize Must Match Filter Radius
fun = @(block) imgaussfilt(block.data, 2);
result = blockproc(huge_image, [512 512], fun, ...
'BorderSize', [6 6], ... % 3*sigma for Gaussian
'TrimBorder', true, ...
'UseParallel', true);
Rule 6: DICOM Normalization โ Use mat2gray, Not im2double
% im2double assumes standard uint16 range; DICOM data may differ
img_norm = mat2gray(double(dicomread('scan.dcm')));
Template Scripts
Ready-to-use medical imaging pipelines in scripts/:
| Script | Purpose |
|---|
template_mri_preprocessing.m | MRI bias correction, denoising, brain extraction |
template_ct_windowing.m | CT Hounsfield unit windowing for different tissues |
template_cell_counting.m | Automated cell detection and counting pipeline |
template_fluorescence_quantification.m | Multi-channel fluorescence analysis |
template_histology_stain_normalization.m | H&E color deconvolution and normalization |
template_adaptive_thresholding.m | Adaptive thresholding with sensitivity tuning |
template_watershed_segmentation.m | Marker-controlled watershed for touching objects |
template_morphological_cleanup.m | Standard binary mask cleanup pipeline |
template_large_image_blockproc.m | Block processing for large images |
template_edge_detection_pipeline.m | Multi-method edge detection comparison |
Knowledge Cards
Medical Imaging (primary value):
knowledge/cards/medical-mri.md โ MRI preprocessing, tissue segmentation, volumetryknowledge/cards/medical-microscopy.md โ Cell counting, fluorescence, H&E histology
Medical-Specific Patterns:
knowledge/cards/filtering-denoising.md โ Modality-specific denoising gotchasknowledge/cards/segmentation-thresholding.md โ Tissue segmentation thresholdingknowledge/cards/morphology-binary.md โ Medical cleanup recipesknowledge/cards/feature-regions.md โ Advanced regionprops for medical measurement
Integration:
knowledge/cards/deep-learning-segmentation.md โ DL + IPT post-processing patternsknowledge/cards/data-types.md โ Type conversion gotchas for medical formats
Cross-Toolbox Integration
For wavelet-based image processing (multiresolution denoising, fusion), use Wavelet Toolbox functions (wavedec2, wdenoise2) for multiresolution denoising and image fusion.
% Example: Wavelet + IPT fusion for MRI denoising
denoised = wdenoise2(mri, 'DenoisingMethod', 'Bayes'); % Wavelet
cleaned = imopen(denoised, strel('disk', 2)); % IPT morphology
enhanced = adapthisteq(cleaned); % IPT contrast
Verified against MATLAB R2025b
