Artificial Intelligence Program to Predict p53 Mutations in Ulcerative Colitis-Associated Cancer or Dysplasia

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Gastrointestinal pathology

Background:

• Accurate diagnosis of colitis-associated neoplasia is crucial for ulcerative colitis treatment.
• Conventional p53 immunohistochemistry is time-consuming and expensive for pathologists.
• Developing AI tools can streamline pathological diagnosis.

Purpose of the Study:

• To develop a deep learning algorithm for predicting p53 immunohistochemical staining from H&E slides.
• To investigate the efficacy of AI in diagnosing colitis-associated neoplasia.
• To provide a cost-effective and time-efficient diagnostic alternative.

Main Methods:

• A convolutional neural network (CNN) was trained on 25,849 patches from H&E and p53-stained slides.
• Glands were annotated and classified as p53 positive, negative, or null.
• The CNN was trained using 80% of the data, with 10% for validation and 10% for testing.

Main Results:

• The CNN achieved a mean average precision of 0.731-0.754 in classifying p53 staining.
• Prediction accuracy ranged from 0.86-0.91, with high specificity (0.91-0.92).
• The model demonstrated good sensitivity (0.73-0.83) and positive predictive value (0.82-0.89).

Conclusions:

• The trained CNN serves as a viable alternative to conventional p53 immunohistochemistry.
• This AI approach offers an accurate, time-saving, and cost-effective method for diagnosing colitis-associated neoplasia.
• The study highlights the potential of AI in improving pathological diagnostic workflows.