Artificial intelligence aided precise detection of local recurrence on MRI for nasopharyngeal carcinoma: a multicenter cohort study.

Background: MRI is the routine examination to surveil the recurrence of nasopharyngeal carcinoma, but it has relatively lower sensitivity than PET/CT. We aimed to find if artificial intelligence (AI) could be competent pre-inspector for MRI radiologists and whether AI-aided MRI could perform better or even equal to PET/CT. Methods: This multicenter study enrolled 6916 patients from five hospitals between September 2009 and October 2020. A 2.5D convolutional neural network diagnostic model and a nnU-Net contouring model were developed in the training and test cohorts and used to independently predict and visualize the recurrence of patients in the internal and external validation cohorts. We evaluated the area under the ROC curve (AUC) of AI and compared AI with MRI and PET/CT in sensitivity and specificity using the McNemar test. The prospective cohort was randomized into the AI and non-AI groups, and their sensitivity and specificity were compared using the Chi-square test. Findings: The AI model achieved AUCs of 0.92 and 0.88 in the internal and external validation cohorts, corresponding to the sensitivity of 79.5% and 74.3% and specificity of 91.0% and 92.8%. It had comparable sensitivity to MRI (e.g., 74.3% vs. 74.7%, P = 0.89) but lower sensitivity than PET/CT (77.9% vs. 92.0%, P < 0.0001) at the same individual-specificities. The AI model achieved moderate precision with a median dice similarity coefficient of 0.67. AI-aided MRI improved specificity (92.5% vs. 85.0%, P = 0.034), equaled PET/CT in the internal validation subcohort, and increased sensitivity (81.9% vs. 70.8%, P = 0.021) in the external validation subcohort. In the prospective cohort of 1248 patients, the AI group had higher sensitivity than the non-AI group (78.6% vs. 67.3%, P = 0.23), albeit nonsignificant. In future randomized controlled trials, a sample size of 3943 patients in each arm would be required to demonstrate the statistically significant difference. Interpretation: The AI model equaled MRI by expert radiologists, and AI-aided MRI by expert radiologists equaled PET/CT. A larger randomized controlled trial is warranted to demonstrate the AI's benefit sufficiently. Funding: The Sun Yat-sen University Clinical Research 5010 Program (2015020), Guangdong Basic and Applied Basic Research Foundation (2022A1515110356), and Guangzhou Science and Technology Program (2023A04J1788).

Parameterized Gompertz-Guided Morphological AutoEncoder for Predicting Pulmonary Nodule Growth.

The growth rate of pulmonary nodules is a critical clue to the cancerous diagnosis. It is essential to monitor their dynamic progressions during pulmonary nodule management. To facilitate the prosperity of research on nodule growth prediction, we organized and published a temporal dataset called NLSTt with consecutive computed tomography (CT) scans. Based on the self-built dataset, we develop a visual learner to predict the growth for the following CT scan qualitatively and further propose a model to predict the growth rate of pulmonary nodules quantitatively, so that better diagnosis can be achieved with the help of our predicted results. To this end, in this work, we propose a parameterized Gempertz-guided morphological autoencoder (GM-AE) to generate any future-time-span high-quality visual appearances of pulmonary nodules from the baseline CT scan. Specifically, we parameterize a popular mathematical model for tumor growth kinetics, Gompertz, to predict future masses and volumes of pulmonary nodules. Then, we exploit the expected growth rate on the mass and volume to guide decoders generating future shape and texture of pulmonary nodules. We introduce two branches in an autoencoder to encourage shape-aware and textural-aware representation learning and integrate the generated shape into the textural-aware branch to simulate the future morphology of pulmonary nodules. We conduct extensive experiments on the self-built NLSTt dataset to demonstrate the superiority of our GM-AE to its competitive counterparts. Experiment results also reveal the learnable Gompertz function enjoys promising descriptive power in accounting for inter-subject variability of the growth rate for pulmonary nodules. Besides, we evaluate our GM-AE model on an in-house dataset to validate its generalizability and practicality. We make its code publicly available along with the published NLSTt dataset.

Deep learning-based precise prediction and early detection of radiation-induced temporal lobe injury for nasopharyngeal carcinoma.

Background: Radiotherapy is the mainstay of treatment for nasopharyngeal carcinoma. Radiation-induced temporal lobe injury (TLI) can regress or resolve in the early phase, but it is irreversible at a later stage. However, no study has proposed a risk-based follow-up schedule for its early detection. Planning evaluation is difficult when dose-volume histogram (DVH) parameters are similar and optimization is terminated. Methods: This multicenter retrospective study included 6065 patients between 2014 and 2018. A 3D ResNet-based deep learning model was developed in training and validation cohorts and independently tested using concordance index in internal and external test cohorts. Accordingly, the patients were stratified into risk groups, and the model-predicted risks were used to develop risk-based follow-up schedules. The schedule was compared with the Radiation Therapy Oncology Group (RTOG) recommendation (every 3 months during the first 2 years and every 6 months in 3-5 years). Additionally, the model was used to evaluate plans with similar DVH parameters. Findings: Our model achieved concordance indexes of 0.831, 0.818, and 0.804, respectively, which outperformed conventional prediction models (all P < 0.001). The temporal lobes in all the cohorts were stratified into three groups with discrepant TLI-free survival. Personalized follow-up schedules developed for each risk group could detect TLI 1.9 months earlier than the RTOG recommendation. According to a higher median predicted 3-year TLI-free survival (99.25% vs. 99.15%, P < 0.001), the model identified a better plan than previous models. Interpretation: The deep learning model predicted TLI more precisely. The model-determined risk-based follow-up schedule detected the TLI earlier. The planning evaluation was refined because the model identified a better plan with a lower risk of TLI. Funding: The Sun Yat-sen University Clinical Research 5010 Program (2015020), Guangdong Basic and Applied Basic Research Foundation (2022A1515110356), Medical Scientific Research Foundation of Guangdong Province (A2022367), and Guangzhou Science and Technology Program (2023A04J1788).

Benefit of [18F] FDG PET/CT in the diagnosis and salvage treatment of recurrent nasopharyngeal carcinoma.

PURPOSE: To compare PET/CT, MRI and ultrasonography in detecting recurrence of nasopharyngeal carcinoma and identify their benefit in staging, contouring and overall survival (OS). METHODS: Cohort A included 1453 patients with or without histopathology-confirmed local recurrence, while cohort B consisted of 316 patients with 606 histopathology-confirmed lymph nodes to compare the sensitivities and specificities of PET/CT, MRI and ultrasonography using McNemar test. Cohorts C and D consisted of 273 patients from cohort A and 267 patients from cohort B, respectively, to compare the distribution of PET/CT-based and MRI-based rT-stage and rN-stage and the accuracy of rN-stage using McNemar test. Cohort E included 30 random patients from cohort A to evaluate the changes in contouring with or without PET/CT by related-samples T test or Wilcoxon rank test. The OS of 61 rT3-4N0M0 patients staged by PET/CT plus MRI (cohort F) and 67 MRI-staged rT3-4N0M0 patients (cohort G) who underwent similar salvage treatment were compared by log-rank test and Cox regression. RESULTS: PET/CT had similar specificity to MRI but higher sensitivity (93.9% vs. 79.3%, P < 0.001) in detecting local recurrence. PET/CT, MRI and ultrasonography had comparable specificities, but PET/CT had greater sensitivity than MRI (90.9% vs. 67.6%, P < 0.001) and similar sensitivity to ultrasonography in diagnosing lymph nodes. According to PET/CT, more patients were staged rT3-4 (82.8% vs. 68.1%, P < 0.001) or rN + (89.9% vs. 69.3%, P < 0.001), and the rN-stage was more accurate (90.6% vs. 73.8%, P < 0.001). Accordingly, the contours of local recurrence were more precise (median Dice similarity coefficient 0.41 vs. 0.62, P < 0.001) when aided by PET/CT plus MRI. Patients staged by PET/CT plus MRI had a higher 3-year OS than patients staged by MRI alone (85.5% vs. 60.4%, P = 0.006; adjusted HR = 0.34, P = 0.005). CONCLUSION: PET/CT more accurately detected and staged recurrence of nasopharyngeal carcinoma and accordingly complemented MRI, providing benefit in contouring and OS.

Deep learning-based growth prediction for sub-solid pulmonary nodules on CT images.

Background: Estimating the growth of pulmonary sub-solid nodules (SSNs) is crucial to the successful management of them during follow-up periods. The purpose of this study is to (1) investigate the measurement sensitivity of diameter, volume, and mass of SSNs for identifying growth and (2) seek to establish a deep learning-based model to predict the growth of SSNs. Methods: A total of 2,523 patients underwent at least 2-year examination records retrospectively collected with sub-solid nodules. A total of 2,358 patients with 3,120 SSNs from the NLST dataset were randomly divided into training and validation sets. Patients from the Yibicom Health Management Center and Guangdong Provincial People's Hospital were collected as an external test set (165 patients with 213 SSN). Trained models based on LUNA16 and Lndb19 datasets were employed to automatically obtain the diameter, volume, and mass of SSNs. Then, the increase rate in measurements between cancer and non-cancer groups was studied to evaluate the most appropriate way to identify growth-associated lung cancer. Further, according to the selected measurement, all SSNs were classified into two groups: growth and non-growth. Based on the data, the deep learning-based model (SiamModel) and radiomics model were developed and verified. Results: The double time of diameter, volume, and mass were 711 vs. 963 days (P = 0.20), 552 vs. 621 days (P = 0.04) and 488 vs. 623 days (P< 0.001) in the cancer and non-cancer groups, respectively. Our proposed SiamModel performed better than the radiomics model in both the NLST validation set and external test set, with an AUC of 0.858 (95% CI 0.786-0.921) and 0.760 (95% CI 0.646-0.857) in the validation set and 0.862 (95% CI 0.789-0.927) and 0.681 (95% CI 0.506-0.841) in the external test set, respectively. Furthermore, our SiamModel could use the data from first-time CT to predict the growth of SSNs, with an AUC of 0.855 (95% CI 0.793-0.908) in the NLST validation set and 0.821 (95% CI 0.725-0.904) in the external test set. Conclusion: Mass increase rate can reflect more sensitively the growth of SSNs associated with lung cancer than diameter and volume increase rates. A deep learning-based model has a great potential to predict the growth of SSNs.