CoChat

Will Machine Learning Tip the Balance in Breast Cancer Screening?

Open access
Trister, Andrew D., Buist, Diana S. M., Lee, Christoph I. · 2017 · JAMA Oncology
Cited by 92

The DOI matches its Crossref record; 4 of the 4 references it cites were found in a registry or an index.

Abstract

Although multiple randomized clinical trials have demonstrated mortality benefit from routine mammography, population-based breast cancer screening continues to be a highly contentious issue. Digital mammography is the most prevalent breast cancer screening tool, but it is imperfect, with a sensitivity of 84% for detecting breast cancers present at the time of screening. The other 16% are not detected owing to a combination of factors, most notably the human limitation of what radiologists are visually able to identify on mammographic images.1 One in 10 women who undergo mammography screening experience screening failures including false-positive examination results that may lead to unnecessary anxiety, biopsies, surgical excision, and treatment. More recently, critics of screening have suggested that a considerable proportion of screendetected breast cancers constitute overdiagnosis or cases that would not have become clinically relevant during women’s lifetimes.2 These collective harms are felt to outweigh mortality benefits among certain subpopulations of women, especially younger women. The ongoing debate regarding the benefits and harms of mammography screening hinges on the assumption that all clinically relevant information is consistently extracted from the images by interpreting radiologists. Yet, radiologists can rely on only a handful of image features that catch the attention of the human eye. The result is a subjective interpretation, with either identification of a suspicious finding needing further diagnostic workup or a determination that cancer is not present. Overall, radiologists rely heavily on their qualitative visual experience without robust quantitative data analyses of available patientand image-level variables. Thus, the question remains whether radiologists’ mammography interpretation truly captures all relevant and available parameters. With recent advances in computer processing power, the exponential growth of digital health data, and cloud-based data storage capabilities, there is growing excitement around the potential for applying machine learning to improve accuracy in clinical medicine. Machine learning, at the intersection of statistics and computer science, can build complex statistical models from big data to potentially identify new variables and combinations of variables that can predict patient outcomes. Newer machine learning techniques, such as deep learning, have shown the ability to select complex, previously unimagined features from images. Applying these techniques to medical images may lay the groundwork for new quantitative imaging assessments and ultimately applications to precision medicine.3 Furthermore, medical imaging appears ripe for disruptive transformation given already existing capabilities in computer vision, automated digital image feature analysis, large stores of digital images through picture archiving and communication systems, linked electronic medical records, and the binary outcomes of imaging-based screening tests like mammography.4 Digital images can be analyzed at the individual pixel level, converting single mammographic images into millions of individual variables. Given a gold standard to learn from, computer algorithms can cluster these pixel-level variables and identify novel image features associated with clinically relevant breast cancers. In essence, machine learning can combine hundreds of thousands of image-level, patient-level, and tumorlevel variables to develop complex algorithms that could dwarf human breast cancer screening accuracy. While promising, machine learning is still in its infancy with respect to demonstrating utility in cancer screening. There is no guarantee that efforts to merge clinical, genomic, biological, and imaging data will lead to improved cancer screening outcomes. Computer assistance is not a new phenomenon in mammography screening. Since 1998, computer-aided detection (CAD) software has been available for screening mammography interpretation and has been touted as a second reader for radiologists. However, CAD has shown no improvements in interpretive accuracy; its use has led to considerably higher false-positive rates, recall rates, and biopsy rates, without an increase in sensitivity or cancer detection rates.5 More recent successes in machine learning applications to identify high-risk breast cancers has been demonstrated by the C-Path tool, an automated pathologist algorithm.6 Rather than focusing on known histologic features, C-Path used unsupervised machine learning and automated image processing to identify thousands of novel imaging features on digital pathology specimens. These novel image-based features were then used to predict patient survival, with performance that surpassed that of community pathologists. One of the key lessons learned from the C-Path project was the need for unsupervised learning to identify previously undiscovered imaging predictors, as opposed to simply relying on known associations.7 Another lesson learned is that extremely large, high-quality, harmonized, unbiased, and generalizable data sets are needed for training and validation of algorithms, a point highlighted by President Obama’s Precision Medicine Initiative. As part of the White House’s Cancer Moonshot Initiative, which aims to accelerate big data analytics to improve cancer outcomes, we have taken on a large and novel effort to address the promise of machine learning and breast cancer screening. In collaboration with VIEWPOINT

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1001/jamaoncol.2017.0473

Cites

4 of 4 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Cited by

See all 92

Metadata

Authors
Trister, A. D., Buist, D. S. M., & Lee, C. I.
Venue
JAMA Oncology
Published
2017-11-01
Volume
3
Issue
11
Pages
1463
ISSN
2374-2437 (ISSN-L)
OpenAlex
W2611638831
Cited by
92
References
4
FWCI
6.07
Type
journal article
Language
en
Open access
green

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.