Deep-Learning Model May Improve Predictions of Survival and Specific Outcomes in Lung Cancer
A study by Xu et al in Clinical Cancer Research evaluating deep-learning networks that analyze time-series computed tomography (CT) images of patients with locally advanced NSCLC has found these networks can integrate imaging scans at multiple time points to improve clinical outcome predictions. Compared with a standard clinical model utilizing parameters such as stage, gender, age, tumor grade, performance, smoking status, and tumor size, the deep-learning model was more efficient in predicting distant metastasis, disease progression, and local regional recurrence.
Lung cancer remains the leading cause of death among both men and women in the United States. According to the American Cancer Society, more than 228,000 cases of lung cancer will be diagnosed this year, and nearly 143,000 people will die from their disease. Non–small lung cancer (NSCLC), the most common type of lung cancer, accounts for 84% of all lung cancer diagnoses, and affects an estimated 190,000 people each year. Most patients with NSCLC have a 5-year survival rate of approximately 18%, making the need for accurately monitoring treatment response and disease progression after treatment crucial.
Study Methodology
For this analysis, the researchers used two independent cohorts of patients—data set A and data set B. Data set A consisted of 179 patients with stage III NSCLC treated with definitive chemoradiation, with pretreatment and posttreatment computed tomography images at 1, 3, and 6 months’ follow-up for a total of 581 scans. The researchers built deep-learning modules using transfer learning, a combination of convolutional neural networks and recurrent neural networks, and single seed-point tumor localization.
Pathologic response validation was performed on data set B, comprising 89 patients with NSCLC treated with chemoradiation and surgery, for a total of 178 scans.
Results
The researchers found that deep-learning models using time-series scans were significantly predictive of survival and cancer-specific outcomes (progression, distant metastases, and locoregional recurrence). Model performance was enhanced with each additional follow-up scan into the convolutional neural network model (eg, 2-year overall survival: AUC = 0.74, P < .05). The models stratified patients into low- and high-mortality risk groups, which were significantly associated with overall survival [hazard ratio = 6.16, 95% confidence interval = 2.17–17.44; P < .001]. The model also significantly predicted pathologic response in data set B (P = .016).
Translational Significance
“Our research demonstrates that deep-learning models integrating routine imaging scans obtained at multiple time points can improve predictions of survival and cancer-specific outcomes for lung cancer,” said Hugo J.W.L. Aerts, PhD, Director of the Computational and Bioinformatics Laboratory at Dana-Farber Cancer Institute/Harvard Cancer Center and principal investigator of this study, in a statement. “By comparison, a standard clinical model relying on stage, gender, age, tumor grade, performance, smoking status, and tumor size could not reliably predict 2-year survival or treatment response.… [D]eep-learning models that quantitatively track changes in lesions over time may help clinicians tailor treatment plans for individual patients and help stratify patients into different risk groups for clinical trials.”
Dr. Aerts is the corresponding author of this study.
Disclosure: Funding for this study was provided by the National Institutes of Health. For full disclosures of the study authors, visit clincancerres.aacrjournals.org.
The content in this post has not been reviewed by the American Society of Clinical Oncology, Inc. (ASCO®) and does not necessarily reflect the ideas and opinions of ASCO®.
