MiThyCA

Microscopic foci of papillary Thyroid Carcinoma-like nuclear features identification with AI in Whole-Slide Images.

MiThyCA is an advanced computational pathology pipeline designed to assist pathologists in the rapid and automated detection of subcentimeter Papillary Thyroid Carcinoma (PTC) foci within Whole-Slide Images (WSIs).

By utilizing a tandem deep-learning architecture, MiThyCA identifies neoplastic areas and detects "sprinkling" areas—tiny, abrupt nuclear alterations that are often time-consuming to find manually. This tool significantly reduces turnaround times, processing a full slide in as little as 11 seconds.

GitHub stars

Computational Pathology Deep Learning

Tandem AI Architecture

MiThyCA employs a sequential “two-step” approach to maximize both sensitivity and computational efficiency:

  1. Model 1 (Neoplasm Finder): A Convolutional Neural Network (CNN) that screens the entire WSI to identify neoplastic or abnormal tissue regions.
  2. Model 2 (PTC-like Feature Detector): A Vision Transformer (TinyViT) that performs a high-resolution analysis only on the areas identified by Model 1, specifically searching for PTC-like nuclear features (e.g., nuclear clearing, grooves, and pseudoinclusions).

Key Results & Performance

The pipeline was validated on a multi-institutional cohort, demonstrating high reliability across various thyroid lesions including NIFTP, PTC, and metastatic lymph nodes.

  • Clinical Value: Effectively detects the “sprinkling sign” in NIFTP and identifies metastatic micro-foci in lymph nodes that might be overlooked during routine manual screening.

Technical Highlights

Tandem Strategy

Combines the efficiency of CNNs with the high-context attention of Vision Transformers (ViT).

Hardware Agnostic

Optimized to run on standard hospital workstations even without high-end dedicated GPUs.

Explainable Heatmaps

Integrated with QuPath to provide visual heatmaps, keeping the pathologist "in-the-loop."

Core Publications

2025

  1. MiThyCA: A Computational Pathology Pipeline for the Identification of Microscopic Foci of Papillary Thyroid Carcinoma-Like Nuclear Features with AI in Whole-Slide Histological Images
    Leone Bacciu, Mario Urso, Vasco Coelho, and 8 more authors
    Endocrine Pathology, 2025