# Robust Semi-Supervised Learning for Histopathology Images through Self-Supervision Guided Out-of-Distribution Scoring

**Authors:** Nikhil Cherian Kurian, Varsha S, Abhijit Patil, Shashikant Khade, Amit Sethi  
**arXiv:** [2303.09930](https://arxiv.org/abs/2303.09930)  
**Date:** 2023-03-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.09930v1/

## Key points

- The paper proposes a pipeline for open-set supervised learning challenges in digital histology images
- Semi-supervised learning is a promising alternative to supervised learning for medical image analysis when obtaining good quality supervision for medical imaging is difficult
- Semi-SL assumes that the underlying distribution of unaudited data matches that of the few labeled samples, which is often violated in practical settings, particularly in medical images
- The presence of out-of-distribution (OOD) samples in the unlabeled training pool of semi-SL can reduce the efficiency of the algorithm and common preprocessing methods may not be suitable for medical images
- The proposed framework efficiently estimates an OOD score for each unlabelled data point based on self-supervised learning to calibrate the knowledge needed for a subsequent semi-SL framework
- The outlier score derived from the OOD detector is used to modulate sample selection for the subsequent semi-SL stage, ensuring that samples conforming to the distribution of the few labeled samples are more frequently exposed to the subsequent semi-SL framework
- This approach preserves all information in data and results in more robust semi-supervised learning
- The proposed method was tailored specifically for medical images and was demonstrated through extensive studies on two digital pathology datasets: Kather colorectal histology dataset and a dataset derived from TCGA-BRCA whole slide images
- The experiments showed that this approach outperformed other semi-supervised learning frameworks
- In conclusion, this multi-stage pipeline provides an effective solution to address open-set supervised learning challenges in digital histology images by efficiently estimating OOD scores and modulating sample selection during subsequent semi SL stages.

## Summary

This paper proposes a novel pipeline for addressing open-set supervised learning challenges in digital histology images. Semi-supervised learning (semi-SL) is a promising alternative to supervised learning for medical image analysis when obtaining good quality supervision for medical imaging is difficult. However, semi-SL assumes that the underlying distribution of unaudited data matches that of the few labeled samples, which is often violated in practical settings, particularly in medical images. The presence of out-of-distribution (OOD) samples in the unlabeled training pool of semi-SL is inevitable and can reduce the efficiency of the algorithm. Common preprocessing methods to filter out outlier samples may not be suitable for medical images that involve a wide range of anatomical structures and rare morphologies. The proposed framework efficiently estimates an OOD score for each unlabelled data point based on self-supervised learning to calibrate the knowledge needed for a subsequent semi-SL framework. The outlier score derived from the OOD detector is used to modulate sample selection for the subsequent semi-SL stage, ensuring that samples conforming to the distribution of the few labeled samples are more frequently exposed to the subsequent semi-SL framework. This approach preserves all information in data and results in more robust semi-supervised learning. The proposed method was tailored specifically for medical images, which typically have a higher degree of novelty than other types of data. The effectiveness of this approach was demonstrated through extensive studies on two digital pathology datasets: Kather colorectal histology dataset and a dataset derived from TCGA-BRCA whole slide images. The experiments showed that our approach outperformed other semi-supervised learning frameworks, demonstrating its effectiveness. In conclusion, this multi-stage pipeline provides an effective solution to address open-set supervised learning challenges in digital histology images by efficiently estimating OOD scores and modulating sample selection during subsequent semi SL stages. This approach can be applied with any semi SL framework and provides a more robust and effective solution for medical image analysis.