[Paper Review] A Survey on Deep Learning of Small Sample in Biomedical Image Analysis

Pengyi Zhang, Yunxin Zhong+3 more|arXiv (Cornell University)|Aug 1, 2019

AI in cancer detectionComputer Science95 references33 citations

TL;DR

A survey of key small-sample learning (SSL) techniques for biomedical image analysis, organized into five categories, and including model explanation methods to aid clinical decision making.

ABSTRACT

The success of deep learning has been witnessed as a promising technique for computer-aided biomedical image analysis, due to end-to-end learning framework and availability of large-scale labelled samples. However, in many cases of biomedical image analysis, deep learning techniques suffer from the small sample learning (SSL) dilemma caused mainly by lack of annotations. To be more practical for biomedical image analysis, in this paper we survey the key SSL techniques that help relieve the suffering of deep learning by combining with the development of related techniques in computer vision applications. In order to accelerate the clinical usage of biomedical image analysis based on deep learning techniques, we intentionally expand this survey to include the explanation methods for deep models that are important to clinical decision making. We survey the key SSL techniques by dividing them into five categories: (1) explanation techniques, (2) weakly supervised learning techniques, (3) transfer learning techniques, (4) active learning techniques, and (5) miscellaneous techniques involving data augmentation, domain knowledge, traditional shallow methods and attention mechanism. These key techniques are expected to effectively support the application of deep learning in clinical biomedical image analysis, and furtherly improve the analysis performance, especially when large-scale annotated samples are not available. We bulid demos at https://github.com/PengyiZhang/MIADeepSSL.

Motivation & Objective

• Motivate the need for SSL in biomedical image analysis due to limited annotations.
• Summarize the main SSL techniques used to improve deep learning performance with small datasets.
• Organize SSL approaches into five categories to guide practice in clinical settings.
• Highlight explanation methods essential for clinical decision making and model transparency.

Proposed method

• Classify SSL techniques into five categories: explanation techniques, weakly supervised learning, transfer learning, active learning, and miscellaneous techniques.
• Discuss data augmentation, domain knowledge, traditional shallow methods, and attention mechanisms as supporting methods.
• Emphasize the integration of explanation methods to support clinical decision making.
• Build a framework to accelerate clinical usage of deep learning in biomedical imaging.
• Provide demos via an online URL to illustrate concepts.

Experimental results

Research questions

• RQ1 What SSL techniques exist for biomedical image analysis when annotated data are scarce?
• RQ2 How can explanation methods and other auxiliary techniques (augmentation, domain knowledge, shallow methods, attention) support SSL in clinical contexts?
• RQ3 How can SSL approaches be organized to enhance practicality and adoption in biomedical imaging tasks?
• RQ4 What are the practical considerations to accelerate clinical deployment of SSL-based models?

Key findings

• Five categories of SSL techniques are identified and explained: explanation techniques, weakly supervised learning, transfer learning, active learning, and miscellaneous approaches.
• Data augmentation, domain knowledge, traditional shallow methods, and attention mechanisms are highlighted as useful auxiliary tools.
• Explanation methods are emphasized as important for clinical decision making and model transparency.
• The survey aims to accelerate clinical usage by compiling and clarifying SSL strategies for biomedical image analysis.
• Demos are provided at an online URL to illustrate the concepts.