Purpose:

Tumors are continuously evolving biological systems, and medical imaging is uniquely positioned to monitor changes throughout treatment. While qualitatively tracking lesions over space and time may be trivial, the development of clinically relevant, automated radiomics methods that incorporate serial imaging data is far more challenging. In this study, we evaluated deep-learning networks for predicting clinical outcomes through analyzing time-series CT images of locally advanced non-small cell lung cancer (NSCLC) patients.

Methods:

We used two independent cohorts, Dataset A and Dataset B consisting in a total of 268 patients with stage III NSCLC for this analysis. Dataset A consists of 179 stage III NSCLC patients treated with definitive chemoradiation, with pre- and post-treatment CT images at 1, 3, and 6 months follow-up (581 scans). Models were developed using transfer-learning of convolutional neural-networks (CNNs) with recurrent-networks (RNN), using single seed-point tumor localization. Pathologic response validation was performed on an independent Dataset-B, comprising 89 NSCLC patients treated with chemoradiation and surgery (178 scans).

Results:

Deep-learning models using time-series scans were significantly predictive of survival and cancer-specific outcomes (progression, distant metastases and local-regional recurrence). Model performance was enhanced with each additional follow-up scan into the CNN model (e.g. 2-year overall-survival: AUC=0.74, p<0.05). The models stratified patients into low and high mortality risk-groups, which were significantly associated with overall-survival (HR=6.16, 95%CI [2.17,17.44],p<0.001). The model also significantly predicted pathological response in Dataset B (p=0.016).

Conclusion:

We demonstrate that deep-learning can integrate imaging-scans at multiple time-points to improve clinical outcome predictions. The model trained on a separate cohort for high and low risk survival was able to distinguish pathological responders. Non-invasive tracking of the tumor phenotype predicted survival, prognosis and pathological response, which can have potential clinical implications on adaptive and personalized therapy.