Deep reinforcement learning enables adaptive-image augmentation for automated optical inspection of plant rust

This study proposes an adaptive image augmentation scheme using deep reinforcement learning (DRL) to improve the performance of a deep learning-based automated optical inspection system. The study addresses the challenge of inconsistency in the performance of single image augmentation methods. It introduces a DRL algorithm, DQN, to select the most suitable augmentation method for each image. The proposed approach extracts geometric and pixel indicators to form states, and uses DeepLab-v3+ model to verify the augmented images and generate rewards. Image augmentation methods are treated as actions, and the DQN algorithm selects the best methods based on the images and segmentation model. The study demonstrates that the proposed framework outperforms any single image augmentation method and achieves better segmentation performance than other semantic segmentation models. The framework has practical implications for developing more accurate and robust automated optical inspection systems, critical for ensuring product quality in various industries. Future research can explore the generalizability and scalability of the proposed framework to other domains and applications. The code for this application is uploaded at https://github.com/lynnkobe/Adaptive-Image-Augmentation.git.

Establishment and Validation of a Novel MRI Radiomics Feature-Based Prognostic Model to Predict Distant Metastasis in Endemic Nasopharyngeal Carcinoma

Purpose

We aimed to establish a prognostic model based on magnetic resonance imaging (MRI) radiomics features for individual distant metastasis risk prediction in patients with nasopharyngeal carcinoma (NPC).

Methods

Regression analysis was applied to select radiomics features from T1-weighted (T1-w), contrast-enhanced T1-weighted (T1C-w), and T2-weighted (T2-w) MRI scans. All prognostic models were established using a primary cohort of 518 patients with NPC. The prognostic ability of the radiomics, clinical (based on clinical factors), and merged prognostic models (integrating clinical factors with radiomics) were identified using a concordance index (C-index). Models were tested using a validation cohort of 260 NPC patients. Distant metastasis-free survival (DMFS) were calculated by using the Kaplan-Meier method and compared by using the log-rank test.

Results

In the primary cohort, seven radiomics prognostic models showed similar discrimination ability for DMFS to the clinical prognostic model (P=0.070-0.708), while seven merged prognostic models displayed better discrimination ability than the clinical prognostic model or corresponding radiomics prognostic models (all P<0.001). In the validation cohort, the C-indices of seven radiomics prognostic models (0.645-0.722) for DMFS prediction were higher than in the clinical prognostic model (0.552) (P=0.016 or <0.001) or in corresponding merged prognostic models (0.605-0.678) (P=0.297 to 0.857), with T1+T1C prognostic model (based on Radscore combinations of T1 and T1C Radiomics models) showing the highest C-index (0.722). In the decision curve analysis of the validation cohort for all prognostic models, the T1+T1C prognostic model displayed the best performance.

Conclusions

Radiomics models, especially the T1+T1C prognostic model, provided better prognostic ability for DMFS in patients with NPC.