The paper presents the 4th COV19D Competition, organized as part of the DEF-AI-MIA Workshop at the 2024 CVPR Conference. The competition includes two challenges: i) Covid-19 Detection and ii) Covid-19 Domain Adaptation. The data used comes from the COV19-CT-DB database, which contains 7,756 3-D chest CT scans, including 1,661 COVID-19 samples and 6,095 non-COVID-19 samples. Each scan consists of 2-D CT slices, with lengths ranging from 50 to 700 slices. The first challenge involves detecting COVID-19 in CT scans, with the dataset split into training, validation, and test sets. The second challenge focuses on domain adaptation, where participants must adapt models to new data from different hospitals and medical centers. The baseline models used are CNN-RNN architectures, which process 3-D CT scans by first analyzing 2-D slices with a CNN and then using an RNN to process the sequence of slices. The model uses a softmax activation function for classification and includes dropout layers for regularization. The competition evaluates performance using the macro F1 score. The baseline models were trained on the COV19-CT-DB database, with data augmentation techniques such as random rotation and horizontal flip applied. The models were trained on a Tesla V100 GPU with a batch size of 5 and a maximum input length of 700 slices. The results show that the baseline models achieved good performance in both challenges, demonstrating the effectiveness of the CNN-RNN architecture in detecting and adapting to COVID-19 in 3-D chest CT scans. The competition highlights the importance of domain adaptation, explainability, and fairness in AI for medical image analysis.The paper presents the 4th COV19D Competition, organized as part of the DEF-AI-MIA Workshop at the 2024 CVPR Conference. The competition includes two challenges: i) Covid-19 Detection and ii) Covid-19 Domain Adaptation. The data used comes from the COV19-CT-DB database, which contains 7,756 3-D chest CT scans, including 1,661 COVID-19 samples and 6,095 non-COVID-19 samples. Each scan consists of 2-D CT slices, with lengths ranging from 50 to 700 slices. The first challenge involves detecting COVID-19 in CT scans, with the dataset split into training, validation, and test sets. The second challenge focuses on domain adaptation, where participants must adapt models to new data from different hospitals and medical centers. The baseline models used are CNN-RNN architectures, which process 3-D CT scans by first analyzing 2-D slices with a CNN and then using an RNN to process the sequence of slices. The model uses a softmax activation function for classification and includes dropout layers for regularization. The competition evaluates performance using the macro F1 score. The baseline models were trained on the COV19-CT-DB database, with data augmentation techniques such as random rotation and horizontal flip applied. The models were trained on a Tesla V100 GPU with a batch size of 5 and a maximum input length of 700 slices. The results show that the baseline models achieved good performance in both challenges, demonstrating the effectiveness of the CNN-RNN architecture in detecting and adapting to COVID-19 in 3-D chest CT scans. The competition highlights the importance of domain adaptation, explainability, and fairness in AI for medical image analysis.