TL;DR
This paper introduces an AI-powered system that uses advanced imaging techniques (SHG/TPEF) to automatically and quantitatively assess hepatic steatosis in mouse models of Metabolic Dysfunction-Associated Steatotic Liver Disease (MASLD). This approach offers a more precise and objective alternative to traditional manual scoring, paving the way for better preclinical drug development and understanding of MASLD progression.
What Problem Does This Paper Address?
Metabolic Dysfunction-Associated Steatotic Liver Disease (MASLD), previously known as NAFLD, is a growing global health concern characterized by fat accumulation in the liver. Its progression can lead to severe liver damage, including inflammation, fibrosis, cirrhosis, and even cancer. Developing effective treatments requires robust preclinical models and accurate methods to assess disease progression. Current methods for evaluating hepatic steatosis in animal models, primarily manual histological scoring, are semi-quantitative, subjective, and suffer from inter- and intra-observer variability. This limits the precision and reproducibility of research, especially when evaluating the dynamic changes in steatosis during disease progression or in response to therapeutic interventions. The paper aims to address this by developing an automated, objective, and dynamic AI-based system for MASLD steatosis assessment.
Methods
The study leverages a combination of advanced imaging and artificial intelligence.
- Imaging: Second-harmonic generation (SHG) and two-photon-excited fluorescence (TPEF) microscopy were used to image unstained liver tissue. SHG visualizes collagen (fibrosis), while TPEF highlights fat vacuoles (steatosis). This technique allows for the acquisition of detailed structural information without the need for traditional staining.
- AI Analysis: An AI system was developed to analyze the SHG/TPEF images.
- Preprocessing: TPEF images were processed using Otsu's thresholding method to identify empty spaces corresponding to fat vacuoles.
- Feature Extraction: The system extracted various features from these potential fat vacuoles, including intensity, area, circularity, density, and length-to-width ratio. It also considered the surrounding collagen area.
- Classification: A Classification and Regression Trees (CART) model was trained to differentiate fat vacuoles from other cellular structures. This model was chosen for its ability to handle feature data without pre-normalization.
- Quantification: The AI system quantified several parameters related to steatosis, such as the percentage of steatosis (% Area), the number of fat vacuoles (# Bubble), and combinations of size, number, and percentage of fat vacuoles within specific histological regions (central vein, portal tract, perisinusoidal, and lobular area).
Datasets
The study utilized liver tissue samples from six different MASLD mouse models, along with a control group. These models were generated using various combinations of:
- Western Diet (WD)
- High-Fructose Drinking Water (WDF)
- High-Fat Diet (HFD)
- High-Fructose Drinking Water (HFDF)
- Carbon Tetrachloride (CCl4) injections (low-dose, weekly)
Mice were sampled at three time points: 0 weeks (baseline), 8 weeks, and 16 weeks. A total of 5 mice per group per time point were used. The AI model was trained using one tile ROI (200 x 200 µm²) from each of the 75 samples.
Key Insights and Ideas
- AI Outperforms Traditional Scoring: The AI system demonstrated a superior ability to quantify dynamic changes in steatosis compared to the traditional NASH CRN histological scoring method, which showed no significant difference between 8 and 16 weeks in some cases.
- Objective and Reproducible Measurement: The AI approach provides an objective and reproducible method for assessing steatosis, overcoming the limitations of subjective manual scoring and inter-observer variability.
- Unstained Imaging Advantage: Using SHG/TPEF for unstained tissue analysis simplifies the workflow and potentially reduces artifacts associated with staining procedures.
- Dynamic Steatosis Progression: The AI system successfully captured the dynamic progression of steatosis across different MASLD models and time points, revealing subtle changes that might be missed by conventional methods.
- Impact of CCl4: The study observed that CCl4 administration, while inducing steatosis, paradoxically led to a decrease in the number of fat vacuoles per unit area compared to diet-only groups, suggesting a complex interplay of factors.
Key Results
- Increased Steatosis Markers: All MASLD mouse models showed increased body weight, serum lipids, and liver enzyme markers at 8 and 16 weeks compared to baseline.
- AI Quantifies Dynamic Changes: The AI system was able to automatically quantify dynamic changes in the area, number, and size of hepatic steatosis.
- Significant Differences Detected: The AI analysis revealed significant differences in steatosis parameters (e.g., % Area, # Bubbles) between 8 and 16 weeks in most MASLD mouse models, a level of detail not consistently captured by traditional scoring.
- Correlation with Traditional Methods: The AI analysis showed strong correlations with traditional H&E staining methods, validating its accuracy. Specifically, correlations between non-stained AI analysis (% Area and # Bubble) and stained AI analysis (% r_Vacuole Area and #Vacuole Count) were high (0.939 and 0.882, respectively).
Limitations and Future Work
- Male Mice Only: The study exclusively used male mice. Future research should investigate potential sex-specific differences in MASLD progression and AI assessment.
- Physiological Parameters: Water balance and intra-abdominal fat were not directly measured, limiting the ability to fully correlate physiological changes with steatosis.
- Cross-Instrument Consistency: While the AI system is promising, the paper notes that publicly available data on the consistency of such automated platforms across different instruments is limited. Future work should focus on standardizing and validating these platforms.
- Cirrhosis and Advanced Disease: The current study focused on steatosis. Future research could extend the AI analysis to evaluate other aspects of MASLD progression, such as inflammation, ballooning, and fibrosis, and to analyze differences in progression towards cirrhosis.
Summary updated