Application of Artificial Intelligence Techniques to Predict Risk of Recurrence of Breast Cancer: A Systematic Review

Abstract

Breast cancer is the most common disease among women, with over 2.1 million new diagnoses each year worldwide. About 30% of patients initially presenting with early stage disease have a recurrence of cancer within 10 years. Predicting who will have a recurrence and who will not remains challenging, with consequent implications for associated treatment. Artificial intelligence strategies that can predict the risk of recurrence of breast cancer could help breast cancer clinicians avoid ineffective overtreatment. Despite its significance, most breast cancer recurrence datasets are insufficiently large, not publicly available, or imbalanced, making these studies more difficult. This systematic review investigates the role of artificial intelligence in the prediction of breast cancer recurrence. We summarise common techniques, features, training and testing methodologies, metrics, and discuss current challenges relating to implementation in clinical practice. We systematically reviewed works published between 1 January 2011 and 1 November 2021 using the methodology of Kitchenham and Charter. We leveraged Springer, Google Scholar, PubMed, and IEEE search engines. This review found three areas that require further work. First, there is no agreement on artificial intelligence methodologies, feature predictors, or assessment metrics. Second, issues such as sampling strategies, missing data, and class imbalance problems are rarely addressed or discussed. Third, representative datasets for breast cancer recurrence are scarce, which hinders model validation and deployment. We conclude that predicting breast cancer recurrence remains an open problem despite the use of artificial intelligence.

Performance of externally validated machine learning models based on histopathology images for the diagnosis, classification, prognosis, or treatment outcome prediction in female breast cancer: A systematic review

“…In addition, the performance metrics, datasets, ML models, image preprocessing, patch sizes, magnifications, and platforms used for annotating/computational purposes were highly variable. This finding corroborates the lack of standardization on the methodology and reported information described by Fell et al 62 and found in previous systematic reviews, such as those reported by Mazo et al with studies using AI tools to predict breast cancer recurrence, 47 by Gao et al with ML-based breast cancer risk prediction models, 63 by Corti et al with AI algorithms for prediction of treatment outcomes in breast cancer, 46 by Nagendran et al with deep learning algorithms for medical imaging, 16 and by Yu et al with deep learning algorithms with EV for radiologic diagnosis. 48 As explained by other authors, this is a noteworthy limitation of these systematic reviews 48 that impedes from making rigorous comparisons, 63 better understand findings, 64 and limits models' generalizability and their clinical impact.…”

“…Similar to what has been described by several authors before, 46 , 47 , 48 , 63 there is an observed lack of consistency in the methods, performance metrics of the studies explored in this work. Additionally, the EVs utilized for each of the studies are heterogeneous and therefore limit the comparison across all the works and possibly inhibit a deeper understanding of how the results compare.…”

“…This limitation has also been described for other ML models developed for medical purposes 16,17,[22][23][24][25][26][27][28][43][44][45] and for diagnostic or predictive purposes on breast cancer patients specifically. [46][47][48] As previously stated, this may be related to the difficulty in finding appropriate external datasets, 47,48 non-adherence to guidelines that promote EV 11,46,[48][49][50][51] and lack of awareness of their importance. 48 Particularly, DP is not yet widely adopted, and robust datasets with reliable labels and appropriate follow-up information have only become available only until recently.…”

“…20,[52][53][54][55][56] Also, as some hold sensitive information or belong to private companies that cannot or do not want to share their data, they have not been publicly available. 47 This will possibly change with an increased awareness of DP applications and benefits and its widespread adoption in upcoming years. 55,57,58 The reason why more of the included studies validated ML models for diagnosis or classification purposes is unknown.…”

Transparency of Artificial Intelligence in Healthcare: Insights from Professionals in Computing and Healthcare Worldwide

“…Furthermore, establishing the responsibility of AI systems whenever they make mistakes-e.g., due to algorithmic bias-may be difficult, as are the accompanying legal actions [4]. It is difficult for these systems to acquire the trust of healthcare professionals and patients and translate them into clinical practice without significant compromise and preparedness of all stakeholders [4,[25][26][27][28][29][30][31].…”