Table of Contents
Introduction
The number of lung cancer-related deaths has been drastically lowered over the past ten years mainly due to the decline in number of smokers, advances in early detection, and improved treatment options; however, lung cancer is still one of the leading causes of cancer-related deaths among both men and women (Siegel, Miller, & Jemal, 2019). It has been reported that in 2017 more people died from this disease alone than from breast, prostate, colorectal, and brain cancers together (Siegel, Miller, & Jemal, 2020).
It was projected that 228,150 new lung cancer cases (13% males and 13% females affected) and 142,670 lung cancer-related deaths (24% males and 23% females affected) will occur in the U.S. in 2019, while these numbers predicted for 2020 are 228,820 (13% males and 12% females affected) and 135,720 (23% males and 22% females affected), respectively (Siegel et al., 2019; Siegel et al., 2020). Based on these predictions, it can be observed that the number of new cases affecting females will be lower than the number of new cases affecting males, and that there will be fewer lung cancer-related deaths affecting both sexes (Siegel et al., 2019; Siegel et al., 2020).
The overall survival rate for lung cancer at 19% is still pretty low in comparison to other types of cancer (Siegel et al., 2020). In addition, the 5-year survival rate for the metastatic lung cancer patients is very low, and is only 5% (Siegel et al., 2020). This is primarily due to the most patients being diagnosed long after the cancer had already spread to other body parts. If lung cancer is detected in the early phase and patients are treated promptly with the customized treatments the 5-year survival rate is estimated to be 57%, which is significantly higher (Siegel et al., 2020). The majority of the lung cancer patients (85%) are diagnosed with non-small-cell lung carcinoma (NSCLC), which is further categorized as either squamous-cell carcinoma (SCC), adenocarcinoma (ADC), or large-cell carcinoma (LCC) (Zappa & Mousa, 2016).
127 our experts are availableright now
On the other hand, 15% of lung cancer patients are diagnosed with small-cell lung carcinoma (SCLC) (Zappa & Mousa, 2016). Distinguishing between these two subtypes of lung cancer is crucial since each requires quite different management and treatment approaches. In order to detect lung cancer in the early phase, characterize the different types, identify certain lung cancer-associated gene mutations at a faster pace, and choose an appropriate clinical trial or customize a treatment plan for pulmonary oncology patients, Artificial Intelligence (AI) should be implemented (Bi et al., 2019; Coccia, 2020; Rabbani, Kanevsky, Kafi, Chandelier, & Giles, 2018; 2018; Xu et al., 2019a).
Methodology/Search Strategy
The literature that discusses the importance and potential ways of using AI (this includes machine learning (ML), and natural language processing (NLP)) in personalized medicine, particularly in pulmonary oncology was reviewed. Peer-reviewed databases such as PubMed, EMBASE, CINAHL, Google Scholar, COCHRANE library, and other online sources were searched for relevant articles. The following key words were used in article search: “lung cancer,” “pulmonary oncology,” “personalized medicine,” “non-small cell lung cancer (NSCLC),” “artificial intelligence (AI),” “machine learning (ML),” “deep learning (DL),” and “natural language processing (NLP).” The literature search for this thesis was limited to English-language articles published from 2010 to 2020. Any peer-reviewed articles that did not fit the search criteria or duplicate articles were excluded.
An extensive search of PubMed provided a total of 2362 hits, however, only 22 articles were deemed relevant. On the other hand, the search of CINAHL provided fewer hits – only 15, with only five being relevant. EMBASE search provided 538 hits with six articles being relevant and five being a duplicate of articles previously found through PubMed. Lastly, the search of Google Scholar provided the highest number of articles, 35,700, however, only 21 were deemed to be relevant and out of those 19, seven were excluded duplicates previously found through PubMed and EMBASE. Additional databases that were searched included COCHRANE library and IEEE Explore, however, these databases did not provide any hits. The search of Science Direct provided 11 relevant articles, six of which were duplicates.
Discussion
In the past, all lung cancers were considered the same and therefore, all lung cancer patients were treated with the identical treatment plan; however, innovations in healthcare had led to the finding that “one size doesn’t fit all.” Luckily, due to these advancements and move towards personalized medicine, this approach has changed. In recent years, AI has been gaining popularity in healthcare, particularly in personalized medicine and in the oncology field. AI uses computer programing and machines which utilize algorithms to extract and retain large volumes of data, such as clinical, genetic, genomic, and environmental information, and subsequently uses this data to help providers make the best possible decision for their patients (Jiang et al., 2017; Meiliana, Dewi, & Wijaya, 2016).
In order to analyze large volumes of the memorized structured data (for instance imaging, genetic testing data, etc.), ML and DL are currently being used. On the other hand, when analyzing large volumes of the retained unstructured data (for instance clinical notes, data extracted from medical journals, clinical laboratory reports, operative notes, discharge summaries, etc.), NLP is used to convert this data to a structured one, which is afterwards analyzed by ML (Jiang et al., 2017). Because of its capabilities, AI has great potential in pulmonary oncology where it has already provided some favorable results such as more precise identification of cancer-affected area, prediction of outcomes, cancer characterization, and detection of possible metastasis (Coccia, 2020; Wang et al., 2019).
Early cancer detection and accurate cancer type and subtype identification are imperative for better patient outcomes and should increase the life span and improve the quality of life for some of these affected patients. In order to detect lung cancer, standard procedures such as the microscopic examination of tissue slides and pathology image analysis have been in practice for years. However, these methods are not only time-consuming but also provide a lot of variabilities in result interpretation (Wang et al., 2019). To make this process more precise and reduce variations, AI should be applied.
ML is a subfield of AI which uses machine algorithms for extracting, analyzing and interpreting unstructured data, and is used for classification and prediction of information that is not easily perceptible to humans (Bi et al., 2019; Jiang et al., 2017; Rabbani, Kanevsky, Kafi, Chandelier, & Giles, 2018). One way that ML can be utilized in lung cancer diagnosis and treatment is the genomic classification of NSCLC, where microarrays are used to train ML algorithms to identify different lung cancer mutations. This is a vital step since personalized medicine can be of great help in this case, and new treatments could be applied based on the existing mutations (Rabanni et al., 2018).
This can be especially beneficial for patients with ADC since 20% of lung ADCs are caused by mutations such as those of the epidermal growth factor receptor (EGFR) gene or rearrangements of the anaplastic lymphoma kinase (ALK) gene which are already being used in personalized medicine as biomarkers for targeted treatments (Cagle, Raparia, & Portier, 2016; Mascaux, Tomasini, Greillier, & Barlesi, 2017). Being able to use ML to detect these biomarkers is beneficial for lung cancer patients who will be not only able to receive the greatest benefits from targeted treatments but will also receive treatment resulting in lower treatment toxicity (Mascaux et al., 2017). In regard to SCC patients, targeted therapies are unfortunately still not available; however, ML is being used for detecting mutations that could possibly be used as biomarkers for targeted treatments.
Besides being used in lung cancer diagnosis, ML is being implemented in determining personalized treatment plans in lung cancer as well as following the prognosis. When deciding on a personalized treatment plan and patient survival, it is vital to distinguish between the two main sub-types of NSCLCs, SCC and ADC, since patients with different NSCLC sub-types require distinctive treatment approaches and have different survival outcomes (Zappa & Mousa, 2016).
In regard to the treatment plan, this characterization is vital since some chemotherapy treatments have been reported to be more successful in treating SCC while other treatments are more successful in treating ADC (Huang et al., 2016; Shoshan-Barmatz et al., 2017). In addition, ML is used to extract quantitative features, such as cell size and shape, scattering of pixel intensity in the cells and nuclei, and texture of the cells and nuclei, from histological slides, which, in turn, will help with distinguishing between cancerous tissue and surrounding healthy tissue (Yu et al., 2016).
Once extracted, these aforementioned quantitative image features in combination with the patient survival indices can be advantageous in determining patient survival outcomes. When predicting survival outcomes in patients with ADC, texture of the nuclei, Zernike shape decomposition of the nuclei, and Zernike shape decomposition of the cytoplasm are of a great interest, while when predicting survival outcomes in patients with SCC, Zernike shape in the tumor nuclei and cytoplasm are being considered (Yu et al., 2016). Additionally, patient features such as genetic, biochemical, physiological, behavioral, as well as environmental exposure are all taken into consideration when AI and ML are used in predicting the survival outcome of lung cancer patients (Schork, 2019).
DL is a subcategory of machine learning and one area of pulmonary oncology where DL application has great potential relates to the prediction of treatment response for NSCLC patients who were undergoing chemotherapy and radiation treatment. Predictions can be achieved by analyzing time series CT images of this group of patients (Xu et al., 2019b). This can be of great help for pulmonary oncologists who will be able, if necessary, to modify treatments for their patients to make them more personalized, which in turn can improve the outcome of those patients.
NLP is a subcategory of AI which can be used for analyzing unstructured data in pulmonary oncology. Once unstructured data is extracted by computers and machines from clinical notes, clinical laboratory reports, pathology reports, surgical notes, and other clinical sources, NLP is used to convert this data to the structured one which can be then analyzed by ML (Jiang et al., 2017). In pulmonary oncology, NLP is primarily being used in assessing lung cancer progression and patient response to given therapies based on the radiology reports (Kehl et al., 2019). However, NLP is not used as often as ML and DL as NLP use is still a new technology, but with the constant advancements in technology, NLP could provide more significant patient-related data that could be used in personalized medicine. All this evidence demonstrates the potential that ML, DL, and NLP could have in personalized medicine, and especially in pulmonary oncology.
Conclusion
Although my article search included peer-reviewed journals between 2010 and 2020, all of the articles relevant to the subject were published in the last years. This shows that AI application is relatively novel in the healthcare field and has great opportunities to make improvements in healthcare, especially in pulmonary oncology and personalized medicine. Use of AI in pulmonary oncology in combination with digital pathology, molecular testing, and electronic health record (EHR) data, can greatly aid in selection of an appropriate treatment drug. Based on data obtained from various studies, AI in pulmonary oncology and personalized medicine is also creating a breakthrough in lung cancer prediction, early detection, cancer characterization, prediction of patient survival, prediction of treatment response.
One of the obstacles to the wider use of AI in pulmonary oncology, and healthcare in general, refers to the data exchange. Continuous supply of data from new clinical studies is vital for the successful application of AI (Jiang et al., 2017). This in turn will enable pulmonary oncologists and radiologists to suggest best possible treatments for their patients, which will then lead to the best possible patient outcomes. Another hurdle to wider use of AI refers to the high cost of initial implementation with many healthcare centers and oncology clinics not being able to afford this cost (Coccia, 2020).
In addition, lack of training of providers in AI utilization is a problem. If providers are adequately trained and are more aware of the advantages they can get from using AI, they would be more likely to incorporate AI more in their work. Moreover, medical imaging data is growing rapidly, and it is challenging to store and share such a large volume of data (Alexander, McGill, Tarasova, Ferreira, & Zurkiya, 2019). Lastly, data privacy presents a big concern and requires the involvement of policy makers and the entire medical community (Rabanni et al., 2018). If these problems are tackled, AI’s use in healthcare will rise significantly in the future.
