Sir Peter MacCallum Department of Oncology - Theses
Permanent URI for this collection
Search Results
1 - 1 of 1
-
ItemA Comprehensive Machine Learning System for Cancer Classification Experiments( 2023-01)Cancer detection, classification/subtyping, grading, segmentation and prognostication are promising applications of machine learning to oncology which, if successful, would yield substantial clinical benefits. Merely knowing a cancer’s subtype provides a clinician with deep insights into its nature and likely progress, effective treatments regimes, and the patient’s prognosis. The volume of studies which make use of machine learning in medicine, including in oncology, is large and growing. A variety of clinical aspirations are evident from the literature, including: improving patient outcomes by increasing cancer detection and classification accuracy; reducing cancer diagnosis costs, and reducing the time required to diagnose cancer. However, researchers wishing to incorporate machine learning into their research face a high barrier, since it requires specialised data science/machine learning skills over and above biomedical expertise. Further, few software tools are available to assist a researcher wishing to undertake such studies. Researchers typically develop necessary software themselves: a difficult and time consuming prerequisite activity to conducting experiments. In response to these shortcomings, I developed ‘CLASSI’: an experiment pipeline for cancer classification/subtyping using whole slide images or RNA-Seq gene expression data. CLASSI makes it straightforward for researchers to incorporate machine learning into their research for one important class of oncology experiments, viz.: cancer classification and subtyping using oncopathology images or RNA-Seq data. CLASSI advances the field by providing an ‘off-the-shelf’ experiment platform to simplify and automate the conduct of histopathology image and RNA-Seq data machine learning experiments, and demonstrates that ‘machine learning enabled’ experiment pipelines are feasible, supporting the case for the development of other, more broadly scoped, experiment pipelines.