³ÉÈË´óƬ

The main goal of computer-aided diagnosis (CAD) systems in medical image analysis is the prediction of clinically relevant outcomes (e.g. diagnosis, prognosis, pathology) directly from medical images.

Over the past few decades attempts have been made to predict outcomes based on CAD systems. However, these first attempts produced unremarkable prediction accuracy for a few reasons, including lack of computational power, and limited medical images available for modelling and testing the systems. A breakthrough came in the ‘90s via machine learning – a process where computer algorithms use datasets containing a large collection of observations (e.g. medical images) and outcomes (e.g. diagnosis) to automatically build an optimal mathematical model that can take previously unseen observations and predict outcomes. Machine learning cut-through the field of medical image analysis in the late ‘90s, as researchers were able to implement models that could segment medical images and predict clinically relevant outcomes using large datasets containing images and relevant annotations (e.g. diagnosis or segmentations). These machine learning based CAD systems showed promising results, but had residual issues.

Deep learning models are highly complex models that tend to require datasets that are orders of magnitude larger than previous machine learning models.

The field of medical image analysis is currently working on the development of CAD systems modelled with extremely large datasets that follow the deep learning paradigm combined with decision explanation.  These ‘smart’ systems can explain the decisions it makes using visual and textual explanations that are easily accessible to doctors. Preliminary results show that these systems can not only produce accurate clinically relevant outcomes, but can explain the decisions for reaching a particular outcome^[123]. In the near future, it is anticipated these CAD systems will be able to automatically discover new imaging biomarkers associated with clinically relevant outcomes, potentially having a significant impact in medicine.

Footnotes

[1] Carneiro G., Pu L Z C T., Singh R., & Burt A 2020, Deep Learning Uncertainty and Confidence Calibration for the Five-class Polyp Classification from Colonoscopy. Medical Image Analysis,Ìý101653.

[2] Maicas G., Bradley A P., Nascimento J C., Reid I & Carneiro G 2019, Pre and post-hoc diagnosis and interpretation of malignancy from breast DCE-MRI, Medical Image Analysis, 58, 101562.

[3] Gale W., Oakden-Rayner L., Carneiro G., Palmer L J., & Bradley A P 2019, Producing Radiologist-Quality Reports for Interpretable Deep Learning, 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019) (pp. 1275-1279). IEEE.

 (Pages 62-64)  

By Gustavo Carneiro, Professor of the School of Computer Science at the ³ÉÈË´óƬ

Professor of the School of Computer Science at the ³ÉÈË´óƬ, ARC Future Fellow, and the Director of Medical Machine Learning at the Australian Institute of Machine Learning. His main research interest are in the fields of computer vision, medical image analysis and machine learning. 

Discover more about CAD aided diagnosis and  list in his recent podcast.