Early detection of breast cancer is vital to improving patient survival rates. Even though machine learning could help by improving access to testing, there are still many open ethical and legal issues related to AI decision-making.

Cancer is not a single disease but rather a collection of over 200 illnesses which share a combination of underlying environmental and genetic causes. This means that cancer is one of the most complicated diseases to manage and cure. It is therefore great news to hear that in 2021 it is forecast that localized/early detected breast cancer five-year survival rates will be approaching 100%.

SEER Stage2020 Breast Cancer 5-year Relative Survival Rate
All SEER stages combined90%

Survival rates have drastically increased since the 1970s (from 76%). This has been due to a number of advances in primary research and care management:

  • HER2-Directed Therapies – by exploiting a protein that cancers often express doctors are able to directly target cancer cells.
  • Gene Expression Testing – using gene testing to detect the presence of genes associated with breast cancer.
  • Hormonal Therapy – therapy can be used to prevent oestrogen from binding to the oestrogen receptor.
  • Less-Extensive Surgery.
  • Exercising and Maintaining a Healthy Weight.

In the future machine learning could be added to this list.

Machine learning and the ethics of AI in cancer care

Machine learning and big data are not new to drug discovery; both techniques have been used in research to 1) allow scientists to analyse large data sets to find novel correlations between genetics, environment and risk factors, and 2) assist in the automation of genome sequencing.

BenchSci alone publishes a list of over 230 start-ups using AI in drug discovery.

AI and robotic experimentation

Liverpool researchers build robot scientist that has already discovered a new catalyst
University of Liverpool robot

A new area of research is in automated experimentation. In July 2020 the University of Liverpool designed a robot that could automatically run experiments 24-7.

“The robot independently carries out all tasks in the experiment such as weighing out solids, dispensing liquids, removing air from the vessel, running the catalytic reaction, and quantifying the reaction products. The robot’s brain uses a search algorithm to navigate a 10-dimensional space of more than 98 million candidate experiments, deciding the best experiment to do next based on the outcomes of the previous ones. By doing this, it autonomously discovered a catalyst that is six times more active, with no additional guidance from the research team”


AI and patients

It has already been proven that AI can assist in the research process; the next step is to enable AI to be more actively deployed nearer to patients. This step is vital as early detection is a significant factor in improving patient outcomes. Any technology that can make testing cheaper, more accessible and widespread will be of great benefit. Currently examinations and mammographic scans are expensive and prioritized according to risk profiling. Whilst this makes logical sense not everyone who ends up with cancer falls into a preidentified risk category.

Using AI to assist the medical decision-making process has come a long way in the past two years. In 2018 the Asian Pacific Journal of Cancer Prevention published results that showed that artificial neural networks (ANNs) could identify breast cancer with a sensitivity of 82% and specificity of 90% based on mammographic scans and patient data. In 2020 these results were improved by researchers from Google Health, DeepMind, Imperial College London, the NHS and Northwestern University. The new model performed with similar accuracy to expert radiologists. The study found that if deployed, the model could reduce the workload of a second reader by 88%.

In the future, AI could assist radiologists and doctors in all stages of the care process, everything from diagnosis and risk calculation to clinical decision-support, by acting as a second pair of trained eyes and an opinion giver. Doctors would therefore be empowered to see more patients and provide faster and more accurate results. The aim should be for more people to be seen sooner and more often!

Additionally, research should be conducted to discover whether some patients would prefer automated machine-testing for breast, cervix, prostate and testicular cancer due to the personal nature of the examination, but this brings in to question the ethics of machine-led/only diagnostics.

Ethics of fully automated diagnostics and treatment

Currently, due to legal and ethical reasons, AI cannot fully automate diagnostics or treatment, but can only be used to augment doctors’ and radiologists’ work. Some of the main issues with AI-first decision-making processes are:

  1. Where does liability lie if a “mistake” is made – is it with the doctor, manufacturer, data scientists, or with the hospital who made the purchasing decision?
  2. If a mistake was unfortunately made, how would the patient prove that there was negligence, bias or error? It might be expensive and complicated for them to prove the algorithm’s effectiveness or bias.
  3. Who would insure against a mistake?

The ethics and insurability of AI is still an evolving area which governments should take an active lead in. The European Union and USA have started publishing initial guidelines. In 2018, the EU released guidelines for the lawful, ethical and robust development of AI.

Geographic distribution of issuers of ethical AI guidelines by number of
documents released. Anna Jobin, Marcello Ienca, Effy Vayena https://arxiv.org/ftp/arxiv/papers/1906/1906.11668.pdf

Additionally, academics have been researching the risks, drivers and potential solutions:

Summary of drivers, risks, solutions and desired outcomes for using AI for breast cancer care. https://doi.org/10.1016/j.breast.2019.10.001

Future areas of improvement

Clinical data highlights two areas that specifically need research and improvement; these are the discrepancy between different races/ethnicities, and knowledge of breast cancer in men.

Differences in race and ethnicity survival rates

In their latest report the American Cancer Society reported a discrepancy between the survival rates of African Americans vs white and Asian/Pacific patients. The hypothesis is that this is partly due to later detection and socio-economic factors. This is an area where AI could assist by making access to treatment more affordable and ubiquitous.

Trends in Female Breast Cancer Death Rates by race/ethnicity
Data from: www.cancer.org

Male breast cancer

Breast cancer in men accounts for around 1% of US patients. Unfortunately there has been a 20% increase since 1975, and this can be partially linked to increased obesity. Additionally, men are more likely to be diagnosed with advanced breast cancer. The American Cancer Society attributes this mostly to a lack of awareness. Potentially, a combination of increased education and automated testing could lead to earlier detection of breast cancer. Introducing automated testing at medical facilities could reduce the embarrassment of men asking for their breast to be checked for signs of cancer.

Breast Cancer Awareness Month

Sadly, not all cancers are detected early enough to benefit from the advancement in treatments. I would like to therefore bring attention to Breast Cancer Awareness Month. Please take a look at https://www.wearitpink.org/ to find out more about fundraising opportunities and how they are working to promote breast cancer awareness.

References and Links

Recommended Posts