DeepPDT-Net: predicting the outcome of photodynamic therapy for chronic central serous chorioretinopathy using two-stage multimodal transfer learning

Abstract

Central serous chorioretinopathy (CSC), characterized by serous detachment of the macular retina, can cause permanent vision loss in the chronic course. Chronic CSC is generally treated with photodynamic therapy (PDT), which is costly and quite invasive, and the results are unpredictable. In a retrospective case–control study design, we developed a two-stage deep learning model to predict 1-year outcome of PDT using initial multimodal clinical data. The training dataset included 166 eyes with chronic CSC and an additional learning dataset containing 745 healthy control eyes. A pre-trained ResNet50-based convolutional neural network was first trained with normal fundus photographs (FPs) to detect CSC and then adapted to predict CSC treatability through transfer learning. The domain-specific ResNet50 successfully predicted treatable and refractory CSC (accuracy, 83.9%). Then other multimodal clinical data were integrated with the FP deep features using XGBoost.The final combined model (DeepPDT-Net) outperformed the domain-specific ResNet50 (accuracy, 88.0%). The FP deep features had the greatest impact on DeepPDT-Net performance, followed by central foveal thickness and age. In conclusion, DeepPDT-Net could solve the PDT outcome prediction task challenging even to retinal specialists. This two-stage strategy, adopting transfer learning and concatenating multimodal data, can overcome the clinical prediction obstacles arising from insufficient datasets.

Laser Treatment of Central Serous Chorioretinopathy – An Update

“…Interestingly, despite overall thickness reduction, the choroidal vascular index is not affected by the procedure. Good responders to PDT have higher choroidal thickness compared to nonresponders 69 , 97 , 98 . This finding is also reflected in a study by Arrigo et al, which differentiate CSCR forms into RPE and choroidal subtypes 99 .…”

MTRA-CNN: A Multi-Scale Transfer Learning Framework for Glaucoma Classification in Retinal Fundus Images

“…Transfer learning method Field of research [17] Simple deep learning network on ImageNet Transfer learning generalizability test for multiple datasets in the image classification domain [18] Xception network pretrain on ImageNet Fingerprint recognition [19] Combining the Xception network pre-trained on imagenet with the designed function block Diagnosis of rectal cancer [20] Different pre-training methods on ImageNet Diabetic retinopathy fundus image classification [21] Pre-trained deep learning networks on the same type of dataset Pediatric pneumonia diagnosis [7] Two-stage transfer learning strategy Central serous chorioretinopathy (CSC) images classification.…”

“…(3) Medium size dataset of same data type: researchers found that transfer learning on medium size datasets of the same type can effectly enhances the network's ability to extract upper-level features. (4) Two-stage transfer learning: in recent years, researchers have combined above transfer learning methods and proposed two-stage transfer learning algorithms, such as [6][7][8][9].This technique can extract more effective features and obtain more accurate results.…”

Deep Transfer Learning for Ethnically Distinct Populations: Prediction of Refractive Error Using Optical Coherence Tomography

“…In a previous study, a model trained on large-scale genome data was first adapted for a small ethnic group through fine-tuning and additional domain adaptation training [ 9 ]. Thus, a new task with a small adaptation dataset can be trained using pre-trained weights from a large dataset through transfer learning [ 10 ]. It is expected that the spread of AI models for various groups will become possible through intra-study adaptation training using ocular images.…”