Table of Links Abstract Abstract Introduction Methods Results Discussion Conclusions, Acknowledgments, and References Introduction Introduction Methods Methods Results Results Discussion Discussion Conclusions, Acknowledgments, and References Conclusions, Acknowledgments, and References 3. Results 3.1. Subjects and Clinical Outcomes 3.1. Subjects and Clinical Outcomes A total of 918 PE patients (median age 64 years, 52% female) and 3,978 CTPAs were identified, with an average of 4 same-day CTPAs per patient. The INSTITUTION1 dataset included 485 patients, while the pooled INSTITUTION2-INSTITUTION3 dataset included 433 patients. 163 patients were deceased at time of review. Sixty-five patients expired within 30 days of diagnosis, and 31 expired within seven days of diagnosis. Furthermore, 77 patients suffered hemodynamic collapse within seven days of diagnosis. Detailed clinical summary is shown in Table 1. Table 1. 3.2. PE Detection Performance 3.2. PE Detection Performance The PENet backbone achieved an accuracy of 0.985 and AUROC of 0.971 in identifying PE on the overall dataset. Furthermore, it achieved a sensitivity of 0.941, a specificity of 1.000, a precision of 1.000, and a F1 score of 0.985. Class Activation Maps (CAMs) were utilized to visualize the neural network’s regions of interest, extracted from the final convolutional layer of PENet and weighted by learned PE classification layer. Highlighted regions of the window prediction indicate predicted locations of PE (Figure 2). Ultimately, the PENet image-based analysis network effectively captured PE-related features in CTPA. Figure 2 3.3. Overall Survival Prediction Performance 3.3. Overall Survival Prediction Performance DL survival analysis frameworks were based on (a) CTPA imaging data, (b) clinical variables, (c) multimodal prediction incorporating both CTPA data and clinical variables, and (d) multimodal model fused with PESI score, as previously established. Model performance is visualized in Figure 3. As a comparative baseline, PESI was evaluated alone. For both internal and external data sets, the PESI-fused model achieved higher c-indices than PESI alone (Table 2). Following stratification of patients into high- and low-risk groups by the PESI-fused model, Kaplan-Meier analysis revealed significantly different mortality outcomes (p<0.001), shown in Figure 4.[24] Figure 3 Table 2 Figure 4 3.4. Short-term Survival Prediction Performance 3.4. Short-term Survival Prediction Performance As PESI estimates risk of 30-day mortality, short-term survival was compared by truncating time-to-event labels at a 30-day maximum.[5] PESI demonstrated greater performance in predicting short-term PE survival compared to long-term (Table 3). However, multimodal and PESI-fused models still exhibited significant performance improvement over PESI in short-term survival prediction. Table 3 3.5. Feature Importance 3.5. Feature Importance For the clinical survival prediction model, a summary of the predictive ability of each clinical feature and respective feature importance within the model is illustrated in Figure 5. Predictive ability measured each variable’s contribution to model performance, while feature importance was determined through coefficients of feature selection with the learning-able neuron weights. Age and history of cancer had the greatest predictive ability, while history of cancer had the greatest feature importance. Figure 5 3.6. Multimodal Improvement 3.6. Multimodal Improvement The multimodal learning framework integrated survival characteristics from multiple modalities- to analyze the individual contributions of each modality, as well as the contribution of PESI to the PESI-fused model, we used NRI to evaluate the accuracy improvement achieved by incorporating each.[20] Predicted risk probabilities were binarized with a threshold of 0.7 to obtain predicted mortality categories. We evaluated NRI by calculating risk scores between (a) imaging and multimodal, (b) clinical and multimodal, and (c) multimodal and PESI-fused models, represented as +Clinical, +Imaging, and +PESI, respectively (Table 4). The positive values of +Clinical and +Imaging indicate that both clinical and imaging data contributed to improved predictive performance of the multimodal framework. Table 4 3.7. RV Dysfunction and PE Risk Classification 3.7. RV Dysfunction and PE Risk Classification We identified RV dysfunction as a risk factor in 16 out of 433 patients in the external test set. We visualized the positioning of the 16 patients in Figure 6a, with the multimodal survival framework identifying 68.8% of RV dysfunction patients as high-risk. The multimodal survival prediction model also demonstrated a high correlation between high-risk identification and mortality, as shown in Figure 6b. Fifty-five of the 65 mortality patients were predicted as high-risk, yielding a mortality classification accuracy of 84.6%. Figure 6a Figure 6b Authors: (1) Zhusi Zhong, BS, a Co-first authors from Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA, Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA, and School of Electronic Engineering, Xidian University, Xi’an 710071, China; (2) Helen Zhang, BS, a Co-first authors from Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (3) Fayez H. Fayad, BA, a Co-first authors from Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (4) Andrew C. Lancaster, BS, Department of Radiology and Radiological Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, 21205, USA and Johns Hopkins University School of Medicine, Baltimore, MD, 21205, USA; (5) John Sollee, BS, Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (6) Shreyas Kulkarni, BS, Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (7) Cheng Ting Lin, MD, Department of Radiology and Radiological Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, 21205, USA; (8) Jie Li, PhD, School of Electronic Engineering, Xidian University, Xi’an 710071, China; (9) Xinbo Gao, PhD, School of Electronic Engineering, Xidian University, Xi’an 710071, China; (10) Scott Collins, Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (11) Colin Greineder, MD, Department of Pharmacology, Medical School, University of Michigan, Ann Arbor, MI, 48109, USA; (12) Sun H. Ahn, MD, Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (13) Harrison X. Bai, MD, Department of Radiology and Radiological Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, 21205, USA; (14) Zhicheng Jiao, PhD, Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (15) Michael K. Atalay, MD, PhD, Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA. Authors: Authors (1) Zhusi Zhong, BS, a Co-first authors from Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA, Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA, and School of Electronic Engineering, Xidian University, Xi’an 710071, China; (2) Helen Zhang, BS, a Co-first authors from Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (3) Fayez H. Fayad, BA, a Co-first authors from Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (4) Andrew C. Lancaster, BS, Department of Radiology and Radiological Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, 21205, USA and Johns Hopkins University School of Medicine, Baltimore, MD, 21205, USA; (5) John Sollee, BS, Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (6) Shreyas Kulkarni, BS, Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (7) Cheng Ting Lin, MD, Department of Radiology and Radiological Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, 21205, USA; (8) Jie Li, PhD, School of Electronic Engineering, Xidian University, Xi’an 710071, China; (9) Xinbo Gao, PhD, School of Electronic Engineering, Xidian University, Xi’an 710071, China; (10) Scott Collins, Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (11) Colin Greineder, MD, Department of Pharmacology, Medical School, University of Michigan, Ann Arbor, MI, 48109, USA; (12) Sun H. Ahn, MD, Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (13) Harrison X. Bai, MD, Department of Radiology and Radiological Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, 21205, USA; (14) Zhicheng Jiao, PhD, Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA; (15) Michael K. Atalay, MD, PhD, Department of Diagnostic Radiology, Rhode Island Hospital, Providence, RI, 02903, USA and Warren Alpert Medical School of Brown University, Providence, RI, 02903, USA. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv available on arxiv
