From Pixel to Slide image: Polarization Modality-based Pathological Diagnosis Using Representation Learning

TL;DR

This work introduces a polarization-modality representation-learning framework for thyroid tumor classification, addressing sampling-related diagnostic challenges by integrating pixel-level and slice-level information through a three-stage pipeline: (1) confidence-learning–assisted pixel-level structure recognition, (2) encoder-decoder–based block-wise feature extraction to distill structural information, and (3) attention-based ROI aggregation for final region classification. The approach leverages polarization parameter maps from both stained and unstained slides, enabling near color-equivalent visualization and enabling diagnosis from unstained tissue. Key contributions include robust handling of label noise via confidence learning, a compact block-level representation via pretrained encoders, and an attention mechanism that assigns importance to image blocks for region-level decisions, achieving high AUC scores (malignant ~0.99, benign ~0.93, indeterminate ~0.86). The methodology supports objective, accurate indirect diagnostics and offers practical benefits such as reduced staining requirements and polarization-based pseudo-coloring to assist pathologists in routine practice.

Abstract

Thyroid cancer is the most common endocrine malignancy, and accurately distinguishing between benign and malignant thyroid tumors is crucial for developing effective treatment plans in clinical practice. Pathologically, thyroid tumors pose diagnostic challenges due to improper specimen sampling. In this study, we have designed a three-stage model using representation learning to integrate pixel-level and slice-level annotations for distinguishing thyroid tumors. This structure includes a pathology structure recognition method to predict structures related to thyroid tumors, an encoder-decoder network to extract pixel-level annotation information by learning the feature representations of image blocks, and an attention-based learning mechanism for the final classification task. This mechanism learns the importance of different image blocks in a pathological region, globally considering the information from each block. In the third stage, all information from the image blocks in a region is aggregated using attention mechanisms, followed by classification to determine the category of the region. Experimental results demonstrate that our proposed method can predict microscopic structures more accurately. After color-coding, the method achieves results on unstained pathology slides that approximate the quality of Hematoxylin and eosin staining, reducing the need for stained pathology slides. Furthermore, by leveraging the concept of indirect measurement and extracting polarized features from structures correlated with lesions, the proposed method can also classify samples where membrane structures cannot be obtained through sampling, providing a potential objective and highly accurate indirect diagnostic technique for thyroid tumors.

Figures (10)

• Figure 1: Polarization parameter maps for different slice thicknesses. (a) 4$\upmu$m thick H&E stained slice. (b) 4$\upmu$m thick unstained slice. (c) 6$\upmu$m thick unstained slice. (d) 8$\upmu$m thick unstained slice. (e) 10$\upmu$m thick unstained slice. (f) 12$\upmu$m thick unstained slice.
• Figure 2: The network architecture for representation learning is divided into three stages: the first stage is microstructure recognition, the second stage involves feature extraction, and the third stage focuses on thyroid tumor classification.
• Figure 3: Polarization parameter maps for different slice thicknesses. (a) 4$\upmu$m thick H&E stained slice. (b) 4$\upmu$m thick unstained slice. (c) 6$\upmu$m thick unstained slice. (d) 8$\upmu$m thick unstained slice. (e) 10$\upmu$m thick unstained slice. (f) 12$\upmu$m thick unstained slice.
• Figure 4: Comparison of Microstructure Recognition Results. (a) Without Confidence Learning. (b) With Confidence Learning. (c) Corresponding H&E Image.
• Figure 5: Importance of Polarization Features. Demonstrates the pivotal role played by polarization features in the process of microstructure recognition.