Abnormality Detection of Effluent Dialysate Images on Continuous Ambulatory Peritoneal Dialysis Using Deep Learning

Continuous Ambulatory Peritoneal Dialysis (CAPD) offers independent dialysis for chronic kidney disease (CKD) patients in Indonesia, enhancing quality of life. However, high mortality risks due to complications prompted by negligence and technical errors persist. In this research, an innovative contribution was suggested: designing a system to detect complications related to CAPD based on abnormality detection of effluent dialysate images using deep learning. The images are harnessed as input for the deep learning model, employing transfer learning techniques atop pre-trained models such as InceptionV3, ResNet50, EfficientNetB7, and MobileNetV2. After this, the model is refined by incorporating supplementary layers, and an array of experiments encompassing diverse dataset resampling scenarios and augmentation strategies are conducted. Empirical findings underscore that the optimal model manifests with ResNet50 under precise circumstances, yielding a recall rate of 75.00% and an f1-score of 78.95% on the test dataset.