Digital Pathology Image-Based Assessment Of The Degree Of Tumor Infiltrating Lymphocyte Infiltration And Prediction Of Metastatic Risk In Nasopharyngeal Carcinoma

ObjectiveThe aim of this study was to calculate the degree of tumor infiltrating lymphocyte(TIL)infiltration in nasopharyngeal carcinoma using deep learning algorithms based on digital pathological images,and also to construct a metastasis risk prediction model to distinguish patients’ metastasis risk and analyze its prognostic predictive value in nasopharyngeal carcinoma.MethodsA total of 404 cases of primary non-metastatic nasopharyngeal carcinoma diagnosed in Jiangxi Cancer Hospital and the Cancer Hospital of Sun Yat-sen University were analysed.The degree of TIL infiltration was first calculated using a convolutional neural network VGG(Visual Geometry Group)model,followed by a metastasis risk prediction model constructed by a clustering-constrained-attention multiple-instance learning approach.Finally,the degree of TIL infiltration and the metastasis prediction score were analysed for their ability to differentiate the risk of metastasis and their independent prognostic value.ResultsThe 5-year DMFS(centre 1: 92.1% vs 71.9%,P=0.00091;centre 2: 97.3 vs90.8%%,P=0.012)and OS(centre 1: 88.8% vs 70.3%,P=0.0067;centre 2: 97.4% vs82.4%,P<0.0001)were significantly higher in patients with high TIL infiltration than in patients with low infiltration,which is calculated by the VGG model.The metastasis prediction model constructed for the study was able to identify patients at high and low risk of metastasis well(AUC=0.86),with 5-year DMFS(57.8% vs96.4%,P<0.001)and OS(64.2 vs 88.8%%,P<0.001)significantly worse in high-risk patients than in low-risk patients.Multi-factorial analysis showed both TIL infiltration and metastasis prediction scores for DMFS(TIL infiltration: HR=0.389,95%CI:0.173-0.874,P=0.022;metastasis prediction score: HR=10.440,95% CI:5.599-19.466,P<0.001)and OS(degree of TIL infiltration: HR=0.374,95%CI :0.195-0.718,P=0.003;metastasis prediction score: HR= 2.779,95% CI: 1.760-4.387,P<0.001)as an independent prognostic factor.ConclusionThe study demonstrated that the degree of TIL infiltration assessed by a deep learning approach based on digital pathology images is closely related to the prognosis of nasopharyngeal carcinoma,and the metastasis prediction score derived from the metastasis prediction model can be used as a valid prognostic indicator for metastasis and death.The deep learning model helps to select patients at high risk of metastasis and provides a reference for more accurate individualized treatment in the future.