Auto Machine Learning for Medical Image Analysis by Unifying the Search on Data Augmentation and Neural Architecture
This work addresses inefficiencies in automated machine learning for medical image analysis, offering incremental improvements to existing methods.
The paper tackled the problem of in-domain sampling bias in small-scale medical image datasets, which causes inefficiency in automated data augmentation methods, by proposing Augmented Density Matching and a unified AutoML algorithm for data augmentation and neural architecture search, achieving state-of-the-art performance on the MedMNIST benchmark.
Automated data augmentation, which aims at engineering augmentation policy automatically, recently draw a growing research interest. Many previous auto-augmentation methods utilized a Density Matching strategy by evaluating policies in terms of the test-time augmentation performance. In this paper, we theoretically and empirically demonstrated the inconsistency between the train and validation set of small-scale medical image datasets, referred to as in-domain sampling bias. Next, we demonstrated that the in-domain sampling bias might cause the inefficiency of Density Matching. To address the problem, an improved augmentation search strategy, named Augmented Density Matching, was proposed by randomly sampling policies from a prior distribution for training. Moreover, an efficient automatical machine learning(AutoML) algorithm was proposed by unifying the search on data augmentation and neural architecture. Experimental results indicated that the proposed methods outperformed state-of-the-art approaches on MedMNIST, a pioneering benchmark designed for AutoML in medical image analysis.