Abstract 1775P
Background
Although deep learning algorithms have shown promising results in medical imaging, their application in predicting adverse pathology events (AP) and biochemical recurrence-free survival (bRFS) in prostate cancer patients is relatively lagging. Here we aimed to evaluate the performance of a novel deep learning network, NAFNet, in predicting AP and bRFS based on pre-treatment multiparametric MRI imaging.
Methods
This multicentre study enrolled 514 prostate cancer patients from six tertiary hospitals throughout China from 2017 and 2021. A total of 367 patients from Fudan University Shanghai Cancer Centre with whole-mount histopathology of radical prostatectomy specimens were assigned to the internal set, and cancer lesions were delineated with whole-mount pathology as the reference. The external test set included 147 patients with BCR data from five other institutes. The prediction model (NAFNet-classifier) and integrated nomogram (DL-nomogram) were constructed based on NAFNet. To evaluate the predictive ability of DL-nomogram for AP, we compared DL-nomogram with radiology score (PI-RADS), and clinical score (Cancer of the Prostate Risk Assessment score (CAPRA)). ROC curves and DCA analyses were performed to assess the AP prediction ability of various models, and survival analyses were also made for bRFS.
Results
After training and validation in the internal set, ROC curves in the external test set showed that NAFNet-classifier alone outperformed ResNet50 in predicting AP (AUC:0.799, 95%CI:0.724-0.873 vs. AUC:0.703, 95%CI:0.618-0.787, P=0.013). The DL-nomogram, including the NAFNet-classifier, clinical T stage and biopsy results, showed the highest AUC (0.915, 95% CI: 0.871-0.959) and accuracy (0.850) compared with the PI-RADS and CAPRA scores. Additionally, the DL-nomogram outperformed the CAPRA score with a higher C-index (0.732, P<0.001) in predicting bRFS.
Conclusions
Our newly-developed deep learning network, NAFNet, combined with clinical factors, accurately predicted AP and poor prognosis in prostate cancer patients from preoperative MRI imaging, providing a potential AI tools in medical imaging risk stratification.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
B. Dai.
Funding
The Medical Innovation Research Project of the Science and Technology Commission of Shanghai Municipality (20Y11905000) and the Discipline Leader of Shanghai Municipal Health Commission (2022XD01).
Disclosure
All authors have declared no conflicts of interest.
Resources from the same session
1794P - Prognostic role of metastatic site in patients with de novo metastatic prostate cancer: A population-based analysis in new hormonal agents era
Presenter: Emre Yekeduz
Session: Poster session 14
1795P - China ARCHES: A multicenter phase III randomized double-blind placebo (PBO)-controlled efficacy and safety trial of enzalutamide (ENZA) + androgen deprivation therapy (ADT) vs PBO + ADT in Chinese patients (pts) with metastatic hormone-sensitive prostate cancer (mHSPC)
Presenter: Gongqian Zeng
Session: Poster session 14
1796P - Real-world analysis of metastatic hormone-sensitive prostate cancer: Are randomized clinical trials more trustworthy? Insights from PIONEER, the European network of excellence for big data in prostate cancer
Presenter: Juan Gómez Rivas
Session: Poster session 14
1797P - Application of novel machine learning model in [68Ga] Ga-PSMA-11 PET/CT: Predicting survival in oligometastatic prostate cancer patients
Presenter: Mikaela Dell'Oro
Session: Poster session 14
1800P - Differential tumor gene-expression profiling of patients (pts) with de-novo metastatic castration-sensitive prostate cancer (dn-mCSPC) versus (vs.) mCSPC relapsing after prior localized therapy (PLT-mCSPC)
Presenter: Vinay Mathew Thomas
Session: Poster session 14
1803P - Phase I/II study of bavdegalutamide, a PROTAC androgen receptor (AR) degrader in metastatic castration-resistant prostate cancer (mCRPC): Radiographic progression-free survival (rPFS) in patients (pts) with <italic>AR</italic> ligand-binding domain (LBD) mutations
Presenter: Daniel Petrylak
Session: Poster session 14