Can Artificial Intelligence (AI) Techniques Improve the Detection of Lung Cancer in PET/CT Imaging? A Systematic Review

Abstract

Background: Most cancer-related fatalities worldwide are caused by lung cancer. The best strategy to raise survival rates has been shown to be early diagnosis and preventative interventions. The manual diagnosis of lung cancer, however, places a heavy and time-consuming load on radiologists. Up until now, there has been a pressing need to use more effective and quick methods. Since certain complex AI models or algorithms have been developed, there has been a rise in studies on the use of artificial intelligence in healthcare, particularly in the detection of numerous diseases, including malignancies, through the analysis of biomedical images. Questions about their success in these difficult challenges remain unanswered despite several attempts to overcome these obstacles.

Aims: The potential benefit of artificial intelligence applications in diagnosing lung cancer, one of the most prevalent cancers worldwide, has received very little attention in published studies up to this point. This systematic review set out to evaluate the diagnostic accuracy of artificial intelligence (AI) models for detecting lung cancer based on PET/CT images.

Methods: Medical databases were systematically searched for research that created or evaluated AI models for detecting lung cancer. following full-text reading and article screening. Ten AI models, including DL and ML, had their performance evaluated, and all of them were then integrated for a quantitative study.

Results: Data from six DL studies with 844 participants were extracted, and the combined sensitivity of the trials was 91.3% (95% CI 86.5%-95.9%). Four ML investigations with 710 samples were further analysed, and the results revealed an overall sensitivity of 87.5% (95% CI 79.6%-95.5%).

Conclusion: A vast majority of the studies championed satisfactory performance metrics in terms of sensitivity with respect to simple detection, benign/ malignant classification or differential diagnosis, and more accuracy than manual lung cancer diagnosis. While the included studies exhibit promising results, it is essential to validate these findings on larger population samples to ensure their high reliability and mitigate potential biases. The accruing data exhibits promising outcomes, signifying that ML and DL tools are poised to assume a substantial and influential role in the domain of AI-based lung cancer screening in the imminent future.