Use of AI in cancer screening
Searched over 200M research papers
Research Analysis by Consensus
Consensus Meter
The Use of Artificial Intelligence in Cancer Screening
AI in Breast Cancer Screening
Accuracy and Efficiency in Mammography
Artificial intelligence (AI) has shown potential in enhancing the accuracy and efficiency of breast cancer screening through mammography. A systematic review of test accuracy studies revealed that while AI systems have demonstrated promising results in smaller studies, they generally underperform compared to radiologists in larger, more comprehensive studies. Specifically, 94% of AI systems evaluated were less accurate than a single radiologist, and all were less accurate than a consensus of two or more radiologists. However, another study highlighted that an AI system could reduce false positives and false negatives significantly, outperforming human experts in breast cancer prediction. This AI system also demonstrated the ability to generalize across different populations, maintaining non-inferior performance and reducing the workload of radiologists by 88% when used in a double-reading process.
Methodological Concerns and Evidence Gaps
Despite the promising results, several methodological concerns limit the large-scale adoption of AI in breast cancer screening. Many studies are based on highly selected image datasets with a high proportion of cancers, which may not be representative of the general population. Additionally, there is limited validation in external datasets and potential biases in training data, which necessitate further research to ensure AI models are transferable to large-scale population-based screening.
AI in Lung Cancer Screening
Diagnostic Performance and Workflow Integration
AI has also been explored for lung cancer screening, particularly using low-dose computed tomography (LDCT). A meta-analysis of deep learning algorithms for lung cancer diagnosis reported a pooled sensitivity of 93% and specificity of 68%, indicating high diagnostic accuracy. AI systems can assist in various stages of the lung cancer screening workflow, from reducing radiation doses while maintaining image quality to aiding in the detection and characterization of lung nodules. These systems can act as a second reader, improving the accuracy of nodule detection and reducing the time required for image interpretation.
Enhancing Clinical Decision-Making
Beyond detection, AI can also support clinical decision-making by predicting tumor properties, treatment responses, and patient prognosis. For instance, AI can help classify gene mutations and PD-L1 expression, which are crucial for personalized treatment plans. Integrating AI into clinical workflows can thus enhance the overall management of lung cancer, from early detection to treatment and prognosis prediction.
AI in Cervical Cancer Screening
Diagnostic Accuracy and Promising Techniques
AI applications in cervical cancer screening have shown promising results, particularly with deep learning (DL) methods. Studies have utilized various imaging techniques, including digital colposcopy and cervicography, with AI algorithms like convolutional neural networks (CNN) and ResNet achieving diagnostic accuracies of over 97%. However, further research is needed to validate these findings and ensure their applicability in clinical settings.
AI in Other Cancer Screenings
Oral and Laryngeal Cancer
AI has also been applied to the early detection of oral and laryngeal cancers. In oral cancer, AI techniques have been used with non-invasive diagnostic tools such as telemedicine and fluorescence imaging, showing potential in early diagnosis. Similarly, AI applications in laryngeal cancer encompass various data modalities, including radiomics and genomics, to support screening and prognosis. However, most studies remain at the proof-of-concept level, requiring larger multi-institutional datasets for optimal performance.
Conclusion
AI holds significant promise in enhancing cancer screening across various types, including breast, lung, cervical, oral, and laryngeal cancers. While AI systems have demonstrated high diagnostic accuracy and potential to improve screening efficiency, several methodological concerns and evidence gaps need to be addressed. Further research and validation in diverse, large-scale datasets are essential to ensure the reliable and effective integration of AI into clinical cancer screening programs.
Sources and full results
Most relevant research papers on this topic