How CT Scans Reveal Fibrosis Through Shape and Texture
Imagine if doctors could detect early liver fibrosis—the scarring that leads to cirrhosis and liver failure—by analyzing subtle patterns in routine CT scans that are invisible to the human eye. This isn't science fiction; it's the cutting edge of medical imaging research happening today. Liver diseases affect billions worldwide, with conditions like metabolic dysfunction-associated steatotic liver disease (MASLD) becoming increasingly common 6 .
For decades, the only definitive way to stage fibrosis was through liver biopsy—an invasive procedure that carries risks of pain, bleeding, and even rare fatalities 1 4 . The quest for non-invasive alternatives has led researchers to an unexpected solution: harnessing the power of shape and texture features in CT images.
Hepatic fibrosis represents the liver's wound-healing response to repeated injury. Think of it as the progressive scarring that occurs when the liver attempts to repair itself from damage caused by viruses, toxins, or metabolic issues.
Potentially reversible with early intervention
The critical treatment threshold
Advanced scarring requiring aggressive treatment
Largely irreversible with potential for liver failure
Conventional CT reading focuses on obvious visual cues like surface nodularity or lobe shrinkage—signs that typically appear only in advanced disease.
Collagen Deposition
Architectural Distortion
Texture analysis goes much deeper, quantifying subtle patterns that reflect underlying structural changes. As collagen deposits accumulate, they create microstructural patterns that alter how tissue appears on CT scans in ways too subtle for human vision to detect 1 .
Basic density measurements including mean, kurtosis, and skewness of pixel values.
Relationship between pixel values such as contrast, entropy, and energy.
Geometric properties of structures within the liver tissue.
Identifying optimal features resembles finding the best ingredients for a complex recipe. Researchers typically follow a multi-step process:
Hundreds of candidate features from CT regions
Using machine learning algorithms
Against biopsy-confirmed fibrosis stages
Features that deliver maximum accuracy
A landmark 2025 study demonstrated the power of radiomics in fibrosis detection 2 . The research team designed a rigorous experiment with 169 patients undergoing both liver biopsy and CT imaging.
Patients enrolled
All fibrosis stages confirmed by biopsy
Non-contrast and contrast-enhanced imaging
The experiment yielded several crucial insights about feature performance in fibrosis detection:
| Model Type | Features Used | CT Type | Normalization | AUC |
|---|---|---|---|---|
| Logistic Regression | Maximum, Energy, Kurtosis, Skewness, Small Area High Gray Level Emphasis | Non-Contrast | Gamma (γ=1.5) | 0.783 |
| Support Vector Machine | Not Specified | Non-Contrast | Various | 0.610 |
| Random Forest | Not Specified | Contrast-Enhanced | Various | 0.568 |
| Feature | What It Measures | Significance |
|---|---|---|
| Maximum | Highest density value | Concentrated collagen deposits |
| Energy | Textural uniformity | Organized collagen deposition |
| Kurtosis | "Tailedness" of distribution | Extreme density values |
| Skewness | Distribution asymmetry | Altered tissue composition |
| SAHGLE | Bright, compact areas | Early fibrotic foci |
| Resource Category | Specific Examples | Function in Research |
|---|---|---|
| Imaging Equipment | 64-detector CT scanners (Canon, Siemens, Philips) | Acquire high-resolution liver images with standardized protocols |
| Contrast Agents | Non-ionic agents (Omnipaque) | Enhance vascular and tissue differentiation when needed |
| Analysis Software | Python libraries (NumPy, Scikit-image), INFINITT PACS | Process images and extract texture features |
| Reference Standards | METAVIR scoring system, biopsy guidelines | Provide gold standard for validating imaging findings |
| Machine Learning Frameworks | Logistic regression, SVM, Random Forest | Identify patterns linking texture features to fibrosis stages |
Beyond the technical tools, the most crucial resource in this field is high-quality, biopsy-matched datasets. The development of accurate models requires large collections of CT images with corresponding histological confirmation from liver biopsies 1 2 .
This sophisticated mathematical approach quantifies topological features in binary images, mapping the "connectivity" patterns of liver tissue 7 .
Deep learning approaches are pushing the boundaries even further. Convolutional neural networks (CNNs) can automatically discover relevant features without human guidance, potentially identifying patterns we haven't yet recognized.
The ultimate goal is to create opportunistic screening tools that run automatically during abdominal CT scans performed for any reason 2 . Such systems could alert radiologists to likely fibrosis even when it's not the primary clinical question.
The quest to identify optimal shape and texture features for staging hepatic fibrosis represents more than technical refinement—it heralds a fundamental shift in how we approach liver disease diagnosis.
Catching liver disease at reversible stages
Reducing dependence on invasive biopsies
Expanding screening to resource-limited settings