Cross-Domain Application of Deep Learning Techniques for Improved Diagnostic and Phenotyping Accuracy in Medical and Agricultural Imaging

Deep learning has advanced medical and agricultural imaging by automating tasks like disease diagnosis and crop phenotyping, but existing models are often specialized and limited by diverse imaging conditions. This study develops a hybrid deep learning framework that integrates methodologies from both fields to improve diagnostic and phenotyping accuracy. The model, pre-trained on a general dataset and finetuned with ultrasound images for pancreatic steatosis diagnosis and field images for pepper phenotyping, used transfer learning and domain adaptation to address domain shifts. It achieved 87.5% accuracy, 90.2% sensitivity, and 84.3% specificity for pancreatic steatosis, and a mean Average Precision (mAP) of 0.92 and Intersection over Union (IoU) of 0.88 for pepper phenotyping, outperforming traditional methods. These results highlight the model’s robustness and potential to bridge the gap between medical and agricultural imaging, suggesting a unified solution for diverse applications and setting the stage for future interdisciplinary research.