Authors (@slack): Malak Abdelfattah (@Malak), Nourhan Mahmoud (@NourhanM1),_ _Mohammed Dahab (@m_dahab7) and Zeyad Ashraf (@Zashraf03)
Cancers of the respiratory system, especially lung cancer, are ranked third as per new cases and top with 75% mortality for the last 5 years figure at 2.2 million new cases. Its heterogeneity and complexity make it difficult and cause drug resistance, thus complicating treatment. The management of several instances of lung cancer over time facilitated the collection of information, which was composed of vast resources of various images and sequences aimed at treating lung cancer. However, to access these databases, we have to use Machine Learning (ML).
The early detection of diseases can be done through traditional techniques such as low-dose CT scans. However, the ML algorithms deal with imaging and omics data which are more complex and enable more accurate and effective detection. ML identifies factors from multi-omics data such as genomics, transcriptomics, etc. To classify patients in high or low-risk categories or patients suffering from lung adenocarcinoma (LUAD) and lung squamous carcinoma (LUSC). This enables patients to get the most appropriate treatment to increase the outcomes.
Clinical studies, imaging, and omics data collection are performed. Further, various algorithms such as logistic regression, random forests, and CNNs are applied to work with the information to classify, extract the features, or identify the anomalies. K-fold cross-validation is an inclusion hack in the models compared to conventional ways.
-
Computer-aided diagnosis (CAD) - As computer programs for medical diagnosis have advanced, they have been successful in helping radiologists interpret complicated imaging data, increasing their accuracy and reducing false positives which are so helpful in cancer diagnosis and treatment.
-
Liquid biopsy analysis – is a remedy through the analysis of blood biomarkers to provide early diagnoses and classification of lung cancer sub-types.
-
Treatment personalization – determines the likely response to possible treatments of an individual based on their genetics and associated molecular composition, especially in improving the effectiveness of planned treatment, such as immunotherapy.
-
Lack of well-annotated, easily accessible medical data with which to train models.
-
The relevance of model robustness and explainability to clinical adoption.
-
Development of standardized performance metrics for ML models in lung cancer.
-
There is a high need to develop appropriate integration and batch removal techniques.
Machine learning has grown as one of the most powerful tools in lung cancer diagnosis, treatment decisions, and prognosis prediction. If the current challenges can be overcome and omics data integrated with imaging, ML is very promising to further personalize the treatments for lung cancer and improve outcomes.
Reference
Li, Y., Wu, X., Yang, P., Jiang, G. and Luo, Y. (2022). Machine Learning for Lung Cancer Diagnosis, Treatment, and Prognosis. Genomics, Proteomics & Bioinformatics, 20(5). doi:https://doi.org/10.1016/j.gpb.2022.11.003.