Article Text
PDF
Abstract
Background Substantial progress has been made in using deep learning for cancer detection and diagnosis in medical images.1-10 However, a significant hurdle for clinical translation of current data-driven deep learning models is lack of interpretability, often attributable to a disconnect from the underlying pathobiology. 11-14 Here, we propose a biology-guided deep learning framework in which a multi-task model is trained to simultaneously predict tumor microenvironment (TME) status and treatment outcomes from radiology images.
Methods 348 patients with gastric adenocarcinoma were used as a training cohort to develop a machine learning algorithm combining radiologic imaging, clinicopathologic data, TME class using ImmunoScore of Gastric Cancer and protein expression of periostin, treatment information, and response to treatment. The model with internally validated with two cohorts of 202 and 636 patients in China. The model was further evaluated with an external validation cohort including 125 and 1062 patients at another Chinese academic medical center. The model was further validated in an American academic medical center cohort consisting of 57 patients. A final immunotherapy cohort of 253 patients was evaluated with advanced gastric cancer undergoing treatment with Anti-PD1 therapy.
Results A total of 2,749 patients were evaluated across training, internal validation, and external validation cohorts. Over 51% of patients with Stage II/III disease received adjuvant chemotherapy. All patients in the immunotherapy cohort had Stage IV disease with 94% harboring mismatch repair deficiency or microsatellite instability-high status. Biology-guided deep learning model was able to accurately predict four distinct TME classes based on computed tomography (CT) imaging 0.81 (95% CI: 0.769-0.851), 0.83 (0.780-0.884), 0.83 (0.780-0.884), and 0.84 (0.776-0.904) which were linked to disease free survival (DFS) and overall survival (OS). The benefit of chemotherapy in DFS and OS was associated with deep learning model and TME class 1-3 (p<0.05), but not class 4. Assessing the effect of immunotherapy (figure 1), combined positive score (CPS) with predicted TME class outperformed CPS alone (Area Under the Curve: 0.805, 95% CI 0.75-0.85 vs 0.642 95% CI 0.58-0.70).
Conclusions Machine learning algorithms trained with pathobiology of tumors may successfully predict TME using only CT imaging in gastric cancer. These algorithms may be prognostic for adjuvant chemotherapy. In addition, use of algorithms may improve prognostication of response to PD-1 therapy despite mismatch repair deficiency and microsatellite instability-high status with use of CPS score of PD-1 expression.
Acknowledgements This project would not have been possible without the assistance and collaboration of staff, faculty, and students of four academic medical centers in China and the United States.
References
Rajkomar A, Dean J & Kohane I. Machine Learning in Medicine. The New England journal of medicine. 2019;380:1347–1358.
Rajpurkar P, Chen E, Banerjee O & Topol EJ. AI in health and medicine. Nat Med. 2022;28:31–38.
Bhinder B, Gilvary C, Madhukar NS & Elemento O. Artificial intelligence in cancer research and precision medicine. Cancer Discov. 2021;11:900–915.
Esteva A, et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017;542:115–118.
Ardila D, et al. End-to-end lung cancer screening with three-dimensional deep learning on lowdose chest computed tomography. Nat Med. 2019;25:954–961.
McKinney SM, et al. International evaluation of an AI system for breast cancer screening. Nature. 2020;577:89–94.
Lotter W, et al. Robust breast cancer detection in mammography and digital breast tomosynthesis using an annotation-efficient deep learning approach. Nat Med. 2021;27:244–249.
Yala A, et al. Optimizing risk-based breast cancer screening policies with reinforcement learning. Nat Med. 2022;28:136–143.
Lambin P, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. 2017;14:749.
Wiens J, et al. Do no harm: a roadmap for responsible machine learning for health care. Nat Med. 2019;25:1337–1340.
Quail DF & Joyce JA. Microenvironmental regulation of tumor progression and metastasis. Nat Med. 2013;19:1423–1437.
Bejarano L, Jordao MJC & Joyce JA. Therapeutic Targeting of the Tumor Microenvironment. Cancer Discov 2021;11:933–959.
Junttila MR. & de Sauvage FJ. Influence of tumour micro-environment heterogeneity on therapeutic response. Nature. 2013;501:346–354.
Fridman WH, Zitvogel L, Sautes-Fridman C & Kroemer G. The immune contexture in cancer prognosis and treatment. Nat Rev Clin Oncol. 2017;14:717–734.
Ethics Approval Ethical approval was obtained from the institutional review board of the four participating centers.
Consent Informed consent was waived for this retrospective analysis.
Prediction of objective response with combined positive score and predicted tumor microenvironment