Dylan Zhang

Dylan Zhang

Auto-generated profile, based on publicly available information

Dylan Zhang

Department of Radiology, Duke University Medical Center, 2301 Erwin Rd, Box 3808, Durham, NC 27710

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
18 May 2023 in Clinical Genitourinary Cancer
Verified
Journal Article
Closed access
HMHannah D. McManus
Hannah D. McManus
DZDylan Zhang
Dylan Zhang
FSFides Regina Schwartz
Fides Regina Schwartz
YWYuan Wu
Yuan Wu
JIJordan Infield
Jordan Infield
MHMichael R. Harrison
Michael R. Harrison

14 November 2024 in American Journal of Neuroradiology
Verified
Journal Article
Closed access
PDF file
JZJikai Zhang
Jikai Zhang
DLDominic LaBella
Dominic LaBella
DZDylan Zhang
Dylan Zhang
JHJessica L. Houk
Jessica L. Houk
JRJeffrey D. Rudie
Jeffrey D. Rudie
ECEvan Calabrese
Evan Calabrese
BACKGROUND AND PURPOSE: To develop and evaluate an automated, AI-based, volumetric brain tumor MRI response assessment algorithm on a large cohort of patients treated at a high-volume brain tumor center. MATERIALS AND METHODS: We retrospectively analyzed data from 634 patients treated for glioblastoma at a single brain tumor center over a 5-year period (2017-2021). The mean age was 56 +/-13 years. 372/634 (59%) patients were male, and 262/634 (41%) patients were female. Study data consisted of 3,403 brain MRI exams and corresponding standardized, radiologist-based brain tumor response assessments (BT-RADS). An artificial intelligence (AI)-based brain tumor response assessment algorithm was developed using automated, volumetric tumor segmentation. AI-based response assessments were evaluated for agreement with radiologist-based response assessments and ability to stratify patients by overall survival. Metrics were computed to assess the agreement using BTRADS as the ground-truth, fixed-time point survival analysis was conducted to evaluate the survival stratification, and associated P-values were calculated. RESULTS: For all BT-RADS categories, AI-based response assessments showed moderate agreement with radiologists’ response assessments (F1 = 0.587-0.755). Kaplan-Meier survival analysis revealed statistically worse overall fixed time point survival for patients assessed as image worsening equivalent to RANO progression by human alone compared to by AI alone (log-rank P=0.007). Cox proportional hazard model analysis showed a disadvantage to AI-based assessments for overall survival prediction (P=0.012). CONCLUSIONS: AI-based volumetric glioblastoma MRI response assessment following BT-RADS criteria yielded moderate agreement for replicating human response assessments and slightly worse stratification by overall survival. ABBREVIATIONS: GBM= Glioblastoma; RANO= Response Assessment in Neuro-Oncology; BTRADS= Brain Tumor Reporting and Data System; NLP = Natural Language Processing.

21 October 2026 in Radiology: Artificial Intelligence
Verified
Journal Article
Closed access
MJMohamed Sobhi Jabal
Mohamed Sobhi Jabal
JZJikai Zhang
Jikai Zhang
DLDominic LaBella
Dominic LaBella
JHJessica L. Houk
Jessica L. Houk
DZDylan Zhang
Dylan Zhang
ECEvan Calabrese
Evan Calabrese
The Brain Tumor Reporting and Data System (BT-RADS) standardizes posttreatment MRI response assessment in patients with diffuse gliomas but requires complex integration of imaging trends, medication effects, and radiation timing. An end-to-end multiagent large language model and convolutional neural network system for automated BTRADS classification was developed and evaluated. Five hundred nine consecutive posttreatment glioma MRI examinations (March 2017-July 2021) from a single center were retrospectively evaluated. An extractor agent identified clinical variables from notes using gpt-oss (OpenAI, 20-billion parameters). A scorer agent applied BTRADS decision logic integrating these with quantitative volumetrics. System accuracy was compared against initial assessments, using expert annotation as the reference standard. Four hundred ninety-two examinations were evaluable. Overall system accuracy was 76.0% (374/492) versus 57.5% (283/492) for initial assessments, an absolute difference of +18.5 percentage points ( P < .001). Cohen kappa between the system and the reference standard was 0.71. The system achieved high sensitivity in context-dependent categories (BT-1b; 100%; BT-1a, 92.7%; BT-3a, 87.5%) and moderate sensitivity for threshold-dependent categories (BT-3c, 74.8%; BT-2, 69.2%; BT-4, 69.3%; BT-3b, 57.1%). The BT-4 positive predictive value was 92.9%. These results support further evaluation of this system as a clinical decision-support tool, with local open-weight inference preserving data privacy. ©RSNA, 2026