breast cancer

As we enter into the last week of Breast Cancer Awareness Month, we can reflect on the definite improvement in the early detection of tumours thanks to the use of mammograms and the use of hormonal replacement therapy in post-menopausal women[1]. In fact from 2004 – 2014 the breast cancer incidence rate has for the most part stabilized. However, according to recent studies, there appears to be a decline in the understanding of the disease and its progression.

According to a global study conducted by Avon that included 19,000 respondents, participants were not confident in identifying the symptoms of breast cancer and when questioned, only 2% checked all the correct symptoms. The study identified a lack of knowledge and understanding with regards to breast cancer symptoms and treatment options which included an understanding of lifestyle choices. For reference, the 2009 Cancer Research UK Report[2] has identified early breast cancer symptoms as:

  • Nipple re-positioning, pulling in
  • Pain, hardness, redness, rash
  • Dimpling of the skin
  • Discharge or bleeding
  • Lump or thickening
  • Change in size or shape of breast

Breast Cancer Diagnosis: 

To complicate matters further, breast cancer diagnosis relies on a multitude of factors that usually includes a pathologist’s examination. Due to the heterogeneous nature of the disease, breast cancer can be further divided into subtypes, the importance of these subtypes further dictates the type of treatment a patient would receive. Even when a patient’s breast cancer subtype is identified there is the risk that the treatment may not work in one patient as it will for another due to unidentified molecular mechanisms. As a result, you may have a population of patients with the same breast cancer subtype but a portion of them might not be receptive to the therapy resulting in a dysfunctional treatment process.

Machine Learning as a Predictive Model: 

The use of artificial intelligence and machine learning for scientific research has the oncology community excited for good reason. Google researchers have already developed deep learning algorithms, trained on large data sets to detect cancer spread within the body. Their early results have shown an 89%[3] degree of accuracy in detecting cancer spread (compared to the current 73%) and promising a higher degree of accuracy in the near future. The data sets would typically include information pertaining to the tumour size, spread, cell uniformity, cell size and other measurements indicative of cancer spread. This information would be considered as the “input” and the subsequent score identifying the cancer’s lethality would be the “output”. The process of training these systems with defined data sets (i.e., containing a defined input and output) is known as supervised learning. As such, theoretically, with the appropriate patient data set, we would be able to train systems to detect malignant lesions requiring immediate attention, indicating which patient requires therapy.

MIT’s Computer Science and Artificial Intelligence Lab published a paper on the development of a machine learning model capable of distinguishing between high-risk breast cancer tumors/lesions requiring immediate surgery. The benefits of such a system would eliminate the uncertainty associated with breast cancer diagnosis and prevent patients from receiving unnecessary surgery if the tumor was deemed benign. IBM’s Watson Health AI has also been used to support physician’s decision making process during the diagnosis stage. Acting as an external opinion (based on training provided through literature review), Watson assesses the cancer case alongside the physician and provides additional support/guidance on the individual cases. It is currently being trained to read mammograms for breast cancer diagnosis.

Whether or not these trained machines will be able to accurately assist pathologists in correctly diagnosing patients and potentially identifying appropriate therapy, remains to be seen. For the time being it is important to be cognizant of the symptoms of breast cancer and to consult your physician with regards to scheduling mammogram sessions. Individuals with a history of breast cancer are also highly encouraged to seek consultation. Know the symptoms, get checked.

[1] Canadian Cancer Society: Breast Cancer Statistics (2017). Retrieved from: http://www.cancer.ca/en/cancer-information/cancer-type/breast/statistics/?region=bc

[2] Breast Module of the Cancer Awareness Measure (Breast-CAM). (Updated 2011 Sept 02). Retrieved from: http://www.cancerresearchuk.org/sites/default/files/health_professional_breast_cam_toolkit_09.02.11.pdf

[3] In-Depth: AI in Healthcare- Where we are now and what’s next. (2017 May 19). Retrieved from: http://www.mobihealthnews.com/content/depth-ai-healthcare-where-we-are-now-and-whats-next