Assessing the probability of metastatic mediastinal lymph node involvement in patients with non-small cell lung cancer using convolutional neural networks on chest computed tomography

BACKGROUND: Lung cancer is the second most common cancer worldwide, accounting for approximately 20% of all cancer-related deaths and having a 10% 5-year survival rate for very late-stage cases. For the prevalent non-small cell lung cancer (NSCLC), recent guidelines advise staging based on the 8th edition of the TNM classification, highlighting the importance of mediastinal lymph node involvement. While noninvasive methods are generally accurate, they often lack sensitivity, and invasive methods may not be suitable for all patients. Advances in deep learning present potential in solving such problems. However, most research focuses on algorithm development more than clinical relevance. Moreover, none of them addressed individual lymph node malignancies, limiting comprehensive analysis and interpretability and leaving clinicians without sufficient means to validate the results effectively. AIM: To develop a local data-trained and validated algorithm for segmenting each mediastinal lymph node in chest computed tomography (CT) and assessing the probability of its involvement in metastasis. MATERIALS AND METHODS: Initially, IASLC lymph node stations are segmented, providing a bounding box of the mediastinum for further processing. Next, the image is cropped to this box and passed through a second network to identify and mask all visible lymph nodes. Finally, each detected lymph node is extracted, stacked with its mask, and evaluated by a feed-forward network to determine malignancy probabilities. RESULTS: The pipeline achieved an average recall and object Dice Score of 0.74±0.01 and 0.53±0.26 for the clinically relevant lymph node segmentation task. Further, it recorded a 0.73 ROC AUC for predicting a patient’s N-stage, outperforming traditional size-based criteria. CONCLUSIONS: The proposed algorithm enables new research algorithms to optimize the management of patients with nonenlarged intrathoracic lymph nodes, thus improving the quality of medical care for patients with cancer.