Code-free machine learning for classification of central nervous system histopathology images

Abstract

Machine learning (ML), an application of artificial intelligence, is currently transforming the analysis of biomedical data and specifically of biomedical images including histopathology. The promises of this technology contrast, however, with its currently limited application in routine clinical practice. This discrepancy is in part due to the extent of informatics expertise typically required for implementation of ML. Therefore, we assessed the suitability of 2 publicly accessible code-free ML platforms (Microsoft Custom Vision and Google AutoML), for classification of histopathological images of diagnostic central nervous system tissue samples. When trained with typically 100 to more than 1000 images, both systems were able to perform nontrivial classifications (glioma vs brain metastasis; astrocytoma vs astrocytosis, prediction of 1p/19q co-deletion in IDH-mutant tumors) based on hematoxylin and eosin-stained images with high accuracy (from ∼80% to nearly 100%). External validation of the predicted accuracy and negative control experiments were found to be crucial for verification of the accuracy predicted by the algorithms. Furthermore, we propose a possible diagnostic workflow for pathologists to implement classification of histopathological images based on code-free machine platforms.

Efficient diagnosis of IDH-mutant gliomas: 1p/19qNET assesses 1p/19q codeletion status using weakly-supervised learning

“…The obtained results go beyond linear regression, as they can reach a diagnostic performance that is as highly reliable as that of conventional FISH. It is widely acknowledged that distinguishing between astrocytoma and oligodendroglioma is challenging due to their mixed features, which makes the classification of low-grade glioma extremely difficult 31 – 33 . Despite this difficulty, 1pNET and 19qNET can discriminate oligodendroglioma from astrocytoma, exhibiting high AUCs of 0.921 and 0.927, respectively.…”

Motivation for using data-driven algorithms in research: A review of machine learning solutions for image analysis of micrographs in neuroscience

“…The barrier to include machine learning algorithms efficiently into research workflows decreases further as time progresses with the invention of new tools that require less and less coding experience ( 27 ). While training and designing deep learning models from scratch for specific tasks is a complex and challenging task, there are already a number of pretrained models for many tasks such as cell segmentation and pathology classification available, which can be used out-of-the-box with the appropriate software.…”

“…We present an overview of those solutions as well as a short comparison between them in Table 3 . Moreover, recent advances in code-free machine learning systems enable researchers with no or little coding knowledge background to leverage machine learning resources with so-called “code-free” machine learning platforms ( 27 ). While these will help with widespread adoption, a thorough understanding of the underlying algorithms and background is recommended to ensure the use of these algorithms to their full potential.…”

A comparative study of an on premise AutoML solution for medical image classification

“…While a number of AutoML benchmarking studies for medical imaging have been performed as surveyed above, there may remain some gaps in the literature that we aim to fill in this study. Firstly, the focus of previous benchmarking efforts has often been focused on comparing multiple AutoML tools 10 , 15 , 20 , 21 , and less against a variety of commonly-used deep learning models. We therefore report performance against several popular deep learning architectures—VGG16, InceptionV3, DenseNet201 and ResNet50 22 – 25 —to provide more context about the added value of AutoML over typical use-cases.…”