AI cancer detection: reducing global disparities and what it means for patients

01How AI improves cancer screening accuracy

Artificial intelligence is the capability of computational systems to perform tasks typically associated with human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making. In cancer screening, AI systems analyze medical images, such as mammograms, CT scans, and pathology slides, to identify patterns associated with malignancy that may be subtle or difficult for human observers to detect consistently.

The objective of cancer screening is to detect cancer before symptoms appear, involving various methods such as blood tests, urine tests, DNA tests, and medical imaging. The purpose of screening is early cancer detection, to make the cancer easier to treat and to extend life expectancy. In 2019, cancer was the second leading cause of death globally, making early detection a critical public health priority.

AI improves screening accuracy by reducing both false positives, where healthy tissue is incorrectly flagged as suspicious, and false negatives, where actual cancers are missed. Machine learning models trained on large datasets of annotated medical images can recognize patterns that correlate with cancer at earlier stages than visual inspection alone. The improvement is most pronounced in screening programs where human readers are fatigued or overworked.

02The global disparity in cancer detection

Cancer detection rates vary dramatically across the world, and this variation reflects profound disparities in healthcare access, infrastructure, and resources. High-income countries have established screening programs, trained radiologists, and modern imaging equipment. Low-income countries often lack these basics, with few imaging machines, fewer trained specialists, and limited access to screening programs.

The disparity means that cancers in low-resource settings are often detected at later stages, when treatment is less effective and survival rates are lower. A cancer that would be detected early and treated successfully in a high-income country may be diagnosed at stage three or four in a low-income country, when curative treatment is no longer possible. This disparity is one of the most significant inequities in global health.

The disparity is widening as high-income countries adopt AI-assisted screening while low-income countries struggle to maintain basic screening programs. If AI tools are deployed only in wealthy health systems, the gap in detection and outcomes will grow. Addressing this requires intentional efforts to make AI tools accessible, affordable, and appropriate for low-resource settings.

03What AI can do in low-resource settings

AI can address several of the barriers that limit cancer screening in low-resource settings. The most obvious is the shortage of trained radiologists and pathologists. In many low-income countries, there are fewer than one radiologist per million population, compared to over 100 per million in high-income countries. AI systems can pre-screen images and flag only the most suspicious cases for human review, multiplying the effective capacity of available specialists.

AI can also reduce the cost of screening by enabling the use of lower-cost imaging equipment. If an AI system can extract diagnostic information from a lower-resolution image, then less expensive machines can be used for screening, with AI compensating for the reduced image quality. This makes screening affordable in settings where high-end imaging equipment is prohibitively expensive.

Mobile health platforms can deliver AI-assisted screening to remote communities. Smartphone-based imaging, combined with cloud-based AI analysis, can bring screening to populations that have never had access to cancer detection. This approach has been demonstrated for cervical cancer screening using smartphone colposcopy and for skin cancer screening using smartphone cameras with AI classification.

04The types of cancer AI detects best

AI detection performance varies by cancer type, with the strongest results in cancers that are screened using standardized imaging protocols. Breast cancer screening with mammography has been one of the most studied applications, with AI systems achieving accuracy comparable to or exceeding that of human radiologists in large clinical trials. Lung cancer screening with low-dose CT is another area where AI has demonstrated strong performance.

Skin cancer detection using dermatoscopic images has also shown promising results, with AI systems matching or exceeding board-certified dermatologists in controlled studies. Cervical cancer screening, particularly with visual inspection and colposcopy, has benefited from AI-assisted image analysis that can identify precancerous lesions with high sensitivity.

Performance is more variable for cancers that require complex tissue analysis or that present with less standardized imaging. Colorectal cancer screening with colonoscopy, prostate cancer screening with MRI, and pancreatic cancer detection remain challenging for AI, though progress is being made. The pattern is that AI excels where the imaging protocol is standardized and the visual features of cancer are well-characterized.

05The regulatory and clinical validation challenge

Oncology is a branch of medicine that deals with the study, treatment, diagnosis, and prevention of cancer. A medical professional who practices oncology is an oncologist. The integration of AI into oncology requires regulatory approval and clinical validation, processes that are designed for traditional medical devices and drugs and that can be slow to adapt to software-based tools that evolve rapidly.

Regulatory agencies such as the FDA in the United States and the EMA in Europe have established pathways for AI-based medical devices, but these pathways require evidence of safety and effectiveness from clinical studies. The challenge is that AI systems improve over time as they are exposed to more data, which means that the system that was validated may not be the same as the system that is deployed. Regulators are developing frameworks for continuous validation, but this remains an area of active debate.

Clinical validation is particularly challenging in low-resource settings, where the patient population, disease patterns, and imaging equipment may differ from those in the high-income countries where most AI systems are developed and validated. An AI system that performs well on Western patient data may not perform as well on African or South Asian populations, where genetic, environmental, and healthcare factors differ. Local validation is essential.

06How AI compares to human radiologists

The comparison between AI and human radiologists is not a simple either-or proposition. In controlled studies, AI systems have matched or slightly exceeded the average radiologist in specific tasks such as mammography interpretation and lung nodule detection. However, these studies typically measure sensitivity and specificity in isolation, while clinical practice involves judgment, context, and communication with patients.

The most effective approach appears to be complementary use, where AI serves as a second reader or a triage tool. AI can pre-screen images and flag suspicious cases for human review, allowing radiologists to focus their attention on the most difficult cases. This approach maintains human oversight while improving efficiency and reducing missed diagnoses, particularly in high-volume screening programs where reader fatigue is a known problem.

In settings with severe specialist shortages, AI may need to function with less human oversight. The challenge is calibrating the appropriate level of autonomy for each setting, ensuring that AI systems are used safely even when human expertise is limited. This requires clear protocols, quality monitoring, and mechanisms for escalating uncertain cases to human review when possible.

07What this means for global health equity

The potential of AI to reduce global disparities in cancer detection is significant but not guaranteed. If AI tools are developed, validated, and deployed only in high-income countries, they will widen the existing gap rather than close it. Achieving equity requires intentional investment in AI tools designed for low-resource settings, local clinical validation, and sustainable deployment models.

Several initiatives are working toward this goal. International partnerships are developing AI systems trained on diverse datasets that include patients from low-income countries. Open-source AI models are being adapted for use with lower-cost imaging equipment. Mobile health platforms are being designed to deliver AI-assisted screening in settings without fixed imaging infrastructure.

The long-term implication is that AI could democratize access to high-quality cancer screening, but only if the technology is treated as a global public health tool rather than a commercial product available only to those who can pay. The decisions made in the next few years about how AI cancer detection is developed, validated, and deployed will determine whether it narrows or widens the global health equity gap.