Machine learning for mediastinal staging of operated patients with the diagnosis of lung cancer

Objective: Primary lung cancer is the most common cause of cancer - related death in the world. Accurate and rapid staging might prolong the survival of patients. Although some methods with high sensitivity and specificity are available for the diagnosis of distant metastasis, the specificity of non-invasive methods for the diagnosis of mediastinal lymph node metastasis is low. We conducted a study to determine an artificial intelligence model for the detection of mediastinal lymph node metastasis with high accuracy. Patients and methods: A total 472 histologically proven non-small cell lung cancer patients, who underwent standard cervical mediastinoscopy in Istanbul Faculty of Medicine Thoracic Surgery Department were included in our study. We performed different artificial intelligence classification and deep learning image processing methods using the clinicopathological and radiological features such as age and gender of patients; side, size, histopathological diagnosis and SUVmax of the tumor; SUVmax ratio of tumor and mediastinal lymph nodes; SUVmax, long and short axis of mediastinal lymph nodes and SUVmax of hilar lymph nodes. We evaluated sensitivity, specifity and accuracy rates as well as F1 score of different artificial intelligence methods. Results: Deep learning radiological image processing model failed to detect mediastinal lymph node metastasis. The sensitivity, specificity and accuracy rates of image processing model were as low as 0 %, 100 %, 50 %, respectively. In SPSS Neural Network, the same rates were determined as 71.4 %, 96.8 %, and 92.3 %, respectively. SMOTE applied logistic regression analysis resulted as 92.3 %, 91.2 %, 91.5 % with F1 score of 0.85, respectively. Linear SVM method, which appeared as the most successful method, these values reached as high as 88.5 %, 98.5 % and 95.7%, respectively. Similarly, the greatest F1 score was found to be as 0.92 in the Linear SVM method. Conclusion: An artificial intelligence model created by Linear SVM method might be used safely with high accuracy in the detection mediastinal lymph node metastasis in non-small cell lung carcinoma.