Researchers have proposed a framework for evaluating deep neural networks in detecting and quantifying lymphoma from PET/CT images. The framework addresses gaps in existing literature by incorporating out-of-distribution testing and comparing model performance with expert human annotators. Four deep networks were tested on 611 cases from multi-institutional datasets, covering various lymphoma subtypes and lesion characteristics. The study found that networks perform better on large, intense lesions with higher metabolic activity. However, both humans and networks struggle to accurately detect small and faint lesions. --- Why it matters: This research is important for engineers working in medical imaging because it provides a more comprehensive framework for evaluating deep neural networks in detecting lymphoma from PET/CT images, which can lead to better treatment decisions for patients. Source: https://arxiv.org/abs/2311.09614