Multiple Instance Learning Framework with Masked Hard Instance Mining for Gigapixel Histopathology Image Analysis
Wenhao Tang, Sheng Huang, Heng Fang, Fengtao Zhou, Bo Liu, Qingshan Liu
2025-09-17
Summary
This paper introduces a new method for analyzing really large, detailed images of tissue samples, which is a growing field called Computational Pathology. The goal is to help computers automatically identify and diagnose diseases like cancer from these images.
What's the problem?
When computers analyze these huge tissue images, they often focus on the easy-to-spot areas, like obvious signs of cancer. This means they can miss more subtle, but still important, clues. The existing methods aren't good at learning from the difficult parts of the images, and those difficult parts are actually really important for making accurate diagnoses.
What's the solution?
The researchers developed a new framework called MHIM-MIL. It works by intentionally focusing on the 'hard' areas of the images – the ones the computer struggles with. They use a system where a 'teacher' model identifies tricky spots, then 'masks' the easy spots, forcing a 'student' model to learn from the challenging ones. They also make sure the hard examples they use are diverse and don't just repeat the same information, and they constantly update the teacher to find even harder examples as the student learns.
Why it matters?
This new method is better at accurately diagnosing cancer and predicting how a patient might respond to treatment. It also does this efficiently, meaning it doesn't take a huge amount of computing power. By focusing on the difficult cases, the computer can learn to identify diseases more reliably, which could lead to earlier and more accurate diagnoses and better patient outcomes.
Abstract
Digitizing pathological images into gigapixel Whole Slide Images (WSIs) has opened new avenues for Computational Pathology (CPath). As positive tissue comprises only a small fraction of gigapixel WSIs, existing Multiple Instance Learning (MIL) methods typically focus on identifying salient instances via attention mechanisms. However, this leads to a bias towards easy-to-classify instances while neglecting challenging ones. Recent studies have shown that hard examples are crucial for accurately modeling discriminative boundaries. Applying such an idea at the instance level, we elaborate a novel MIL framework with masked hard instance mining (MHIM-MIL), which utilizes a Siamese structure with a consistency constraint to explore the hard instances. Using a class-aware instance probability, MHIM-MIL employs a momentum teacher to mask salient instances and implicitly mine hard instances for training the student model. To obtain diverse, non-redundant hard instances, we adopt large-scale random masking while utilizing a global recycle network to mitigate the risk of losing key features. Furthermore, the student updates the teacher using an exponential moving average, which identifies new hard instances for subsequent training iterations and stabilizes optimization. Experimental results on cancer diagnosis, subtyping, survival analysis tasks, and 12 benchmarks demonstrate that MHIM-MIL outperforms the latest methods in both performance and efficiency. The code is available at: https://github.com/DearCaat/MHIM-MIL.