Graduating from 'Ashcroft Score': How AI Pathology Solves Variability and Objectivity in Fibrosis Assessment

Lead: A long-standing headache in pulmonary and liver fibrosis research implies the "subjectivity" and "variability" of pathology assessment. "Dr. A said Score 3, but Dr. B said Score 2." Such fluctuations among evaluators not only undermine the reliability of drug efficacy data but also cause the oversight of promising drug candidates (Type II Error). In this article, we propose a paradigm shift from the classic Ashcroft Score to "Pixel-level Quantitative Analysis" using AI (Artificial Intelligence) technology.

Key Takeaways

Structural problems of "variability" in Ashcroft Score

Benefits of quantitative analysis by AI Digital Pathology (Improved S/N ratio)

Latest examples of "Augmented Intelligence" where pathologists and AI collaborate

1. Limitations of Traditional "Scoring"

The Ashcroft Score (0-8 point grade evaluation), most commonly used for evaluating Idiopathic Pulmonary Fibrosis (IPF) models, has structural flaws.

1. Inter/Intra-observer Variability

Even skilled pathologists can have judgments waver depending on the day or the viewing location. Especially in subtle cases like "boundary between Grade 3 and 4," judgment is left to subjectivity.

2. Difficulty of "Global" Assessment

It is difficult to uniformly evaluate the entire lung. Humans tend to unconsciously focus on "parts with strong lesions," causing bias to overestimate (or underestimate) the severity as a whole.

2. 3rd Generation Analysis: Digital Pathology using AI

Currently, advanced image analysis platforms like HALO® (Indica Labs) and QuPath are spreading, advancing the "complete automation and objectification" of fibrosis evaluation.

What is AI Seeing?

AI (Machine Learning / Deep Learning models) captures the entire slide glass as a virtual slide and analyzes it in the following process.

Tissue Classifier: Automatically paints separate areas for background (blank), normal alveoli, bronchi, blood vessels, and "fibrotic lesions".
Pixel Quantification: Calculates as an absolute value, such as "Of the total lung field area of 50mm², the fibrosis area is 12.5mm²," rather than "sort of 3 points."
Detection of Microstructures: Can detect slight thickening of alveolar septa or differences in collagen fiber density that are easily missed by the human eye.

3. Case Study: AI vs Human Eye

Data from an actual comparative study. Drug efficacy evaluation was performed when administering an anti-fibrotic drug (Pirfenidone) to a bleomycin model.

Evaluation Method	P-value (Placebo vs Treated)	Judgment Result
Ashcroft Score (Human)	p = 0.08	Not Significant<br>Variability was large, drug efficacy could not be detected.
AI Image Analysis (Fibrosis Area %)	p = 0.03	Significant<br>Detected subtle differences and proved drug efficacy.

Thus, using AI dramatically improves the S/N Ratio (Signal-to-Noise Ratio) of data, potentially enabling the detection of significant differences with fewer N numbers (= reduction of study costs).

4. "Augmented Intelligence": Coexistence of Pathologist and AI

It is often misunderstood, but AI does not take away the pathologist's job. Rather, it "Augments" it.

Role of AI: As a tireless calculator, it performs area calculations and cell counts of the vast entire tissue.
Role of Pathologist: QC of whether AI is recognizing correctly, and biological interpretation of "Why was such a lesion formed?"

Advanced CROs and research institutions realize robust data packages that can withstand submission to regulatory authorities (PMDA/FDA) by providing hybrid reports of "Quantitative Data by AI" + "Finding Comments by Board Certified Pathologists".

5. Conclusion: "Conviction" with Objective Data

"Experiments are not going well," "Data varies." The cause may be in the "evaluation method," not the drug. Let's stop worrying about subjective scores. Objective numerical data by AI will be the sure compass that guides your drug discovery project to the next phase (clinical trials).