Evaluation of Deep Learning Systems in Medical Diagnosis

Abstract

Deep learning in medical image analysis has significantly improved diagnostic accuracy. However, using commercial solutions such as Google DeepMind Health, IBM Watson Health, and Aidoc is financially demanding, limiting their adoption in many healthcare institutions. In contrast, open-source systems like MONAI, nnU-Net, and DeepHealth Toolkit offer high efficiency in medical image analysis without substantial financial costs. This study evaluates their performance using metrics such as Dice Coefficient, Precision, Recall, and F1-score, comparing them with the results of commercial solutions.