Hierarchical Diagnosis for Adult Diffuse Glioma Based on Deep Learning

Artificial Intelligence Research of Hierarchical Diagnosis for Adult Diffuse Glioma Based on Deep Learning

This is a restrospective study to establish a deep learning model based on multi-parametric magnetic resonance imaging scans to predict Grade, histopathologic type and genotype of adult diffuse Glioma.

Glioma is a common kind of tumor in central nervous system. The pre-operative prediction of grade, histopathologic type and genotype is important for treatment and management of Adult diffuse Glioma patients. Right now, most of the diagnostic prediction models on glioma are based on 2016 WHO central nervous system tumor guideline. The goal of this study is to establish a new deep learning model to predict Grade, histopathologic type and genotype of adult diffuse Glioma. We will recruit 500 patients with pathologically confirmed diagnosis of Glioblastoma, Astrocytoma and Oligodendroglioma who received neurologic surgery in our center. Each subject underwent pre-operative multi-parametric magnetic resonance imaging scans including T1WI, T2WI, T1CE, FLAIR and DWI. Pathologic diagnosis of each patient are available in pathology department. A deep learning based hierarchical diagnosis

Inclusion Criteria:

1. Patients undergoing surgery in Nanjing DrumTower Hospital between 2010.01 and 2022.05 with post-surgery pathological diagnosis of WHO Grade II to IV Adult Diffuse Glioma.
2. Available pre-surgery T1WI, T2WI, T1CE, FLARI and DWI MR sequences
3. No pre-surgery anti-tumor therapy

Exclusion Criteria:

1. Poor image quality
2. Failed image preprocessing
3. Unavailable pathology data