Skip to main content
Home » Oncology » Artificial intelligence – great potential to reduce inefficiencies, errors and cost
Innovations in Oncology 2019

Artificial intelligence – great potential to reduce inefficiencies, errors and cost

iStock / Getty Images Plus / Lordn

Dr Partha Basu, MD, PhD

Head, Screening (SCR) Group, International Agency for Research on Cancer (World Health Organization), Lyon

Cancer detection in women may be more efficient and affordable with the use of artificial intelligence. However, more research is needed to address its real-world applicability and safety.

Cancer early detection requires examination by a trained health provider using different imaging techniques, endoscopic examinations and pathological investigations.

Irrespective of the methodology, the detection of diseases in medical science is often subjective. This is true for women’s cancer and precancer detection and poses two major challenges.

Firstly, a certain number of misdiagnosis is inevitable due to interpretation error. The second and more formidable challenge is to get an adequate number of trained personnel, the need for which is growing exponentially.

Artificial Intelligence (AI) has the potential to automate detection, improve accuracy, enhance efficiency and reduce healthcare costs. A computer processing 250 million images may cost as little as $1,000.

How does AI work to detect cancer?

Image recognition software programmes can be ‘trained’ to detect patterns by processing massive datasets using a variety of artificial neural network (ANN) configurations.

ANN requires digitised inputs in the form of supervised learning, which allows the computer to detect features with increasingly high confidence and ultimately, be able to detect/predict the abnormalities accurately.

ANN can help interpret vital signs, skin changes, ECGs, medical scans, endoscopy images, pathology slides – the possibilities are endless.

AI can act as ‘double reader’ in place of a second radiologist

Mammographies may be used for breast cancer screening and for the detection of disease in symptomatic women.

The assessment of mammograms is prone to both reading error (failure to detect abnormality) and decision error (incorrect interpretation). Mammograms miss up to 30% of the cancers.

Independent reading by two radiologists, which is recommended to reduce error, increases the burden on the radiology services.

ANN has been used to complement a radiologist as a ‘double reader’. AI technology has yet to demonstrate conclusively the capacity to out-perform a trained radiologist.

Nearly 15% of breast biopsies performed because of abnormal mammograms are stamped as ‘high-risk’ conditions, and are essentially benign.

Women undergo breast surgery just because the true malignant potential of such conditions is uncertain. The development of validated AI algorithms is likely to be able to accurately risk stratify such patients based on clinical and laboratory data. This should allow us to avoid unnecessary surgeries.

AI in cervical cancer screening avoids subjective readings

Screening for cervical cancer traditionally used Pap smear. Women with a positive smear are usually referred to colposcopy (visually examines the cervix under light illuminated magnification) to confirm the diagnosis.

Colposcopy is highly subjective. The examination is therefore error-prone and may be improved with use of AI.

An algorithm developed by the US National Cancer Institute identified pre-cancer/cancer using several thousand cervical images. The system achieved greater accuracy than human interpretation of an image or Pap smear.

Colon cancer detection could be improved with AI

Individuals positive on screening test for colon cancer or having suspicious symptoms require colonoscopy to detect polyps (many are precancerous) or cancer.

AI can identify very tiny polyps in the colon during colonoscopy with better accuracy and speed than manual detection by an endoscopist.


Advances in AI development must match the anatomic and pathologic complexities of health and disease.

There are ethical and legal issues around the ownership of responsibility for any harms resulting out of ‘technological failures’ or compromised privacy.

Questions remain about whether health-providers will be made redundant as AI replaces many of the critical decision-making and will medical-care lose the human touch?

Perhaps not. The accuracy and efficiency of AI with appropriate oversight of algorithmic interpretations by a skilled provider may provide the best approach to deliver efficient and equitable future health-care.

Disclaimer:As personnel of the International Agency for Research on Cancer / World Health Organization, the author alone is responsible for the views expressed in this article and he does not necessarily represent the decisions, policy or views of the International Agency for Research on Cancer / World Health Organization.

Next article