Leveraging AI for lung cancer prediction

Extract

[...] Early lung cancer often has no symptoms and can only be detected by medical imaging. Today, according to the National Cancer Institute, a maximum of 150 days can be observed between a patient's first examination and the start of the lung cancer treatment. Survival rates are higher in those diagnosed at an earlier stage, and diagnosed at a younger age. Therefore, MSD should develop an AI based solution aimed at healthcare physicians which predicts cancer risk, development and diagnoses in patients. [...]

[...] Details and Value of Proposed Solution The outcome: Based on a given patient's medical history and exam results, the solution will be able to determine the risk of developing lung cancer. High risk patients will be identified by the tool. Their files will be reviewed by medical professionals. Early lung cancer has no symptoms but as the cancer progresses, most patients experience respiratory problems. If the risk is confirmed, patients will undergo a series of imaging tests to determine the location and extent of any tumors. [...]

[...] A definitive lung cancer diagnosis will require a biopsy of the suspected tumor. With an early identification of patients at risk, the time between diagnosis and treatment will be reduced. Machine learning techniques: The algorithm will be trained with both structured (patient's age, zip code?) and unstructured data (doctor's transcripts, medical imaging?).In order to develop the solution, data scientists will use natural processing language to extract key insights and data (symptoms, profiles, results?) from the medical reports. The solution will also rely on both supervised and unsupervised learning. [...]

[...] The solution will therefore be able to classify and influence cancer screening recommendation outcomes. This solution will be an early triage tool, not a substitute for medical professional concertations or diagnostics. Data availability: To train the algorithms, data such as : anonymised medical patient records, reimbursement structured data from the French government system, pathology profiles, historical diagnostic data from hospitals, imaging studies, and medical exam results will be used. The first iterations of the solution should focus on the most widespread types and forms of lung cancer. [...]

[...] In order to mitigate those risks investment is needed in infrastructure for high-quality and representative data. The solution, and any AI systems, learn from the data on which they are trained, and they can incorporate biases from those data. Therefore, data scientists should ensure that the training data comes from diverse and reliable sources. Finally, medical professionals should be aware of the limits of any AI powered tool and learn how to catch and correct its errors and further develop medical knowledge. [...]