Diabetes foot ulceration (DFU) and rheumatoid arthritis (RA) are two complex diseases that affect many people around the world. In this M.Sc. thesis, I present two approaches to address early detection and prediction challenges for these conditions. For DFU, previous trials have shown the potential of thermograms in detecting increased plantar temperature before DFU, but identifying it early has been challenging due to complex temperature distribution. Hence, I develop a multi-stream, multi-view convolutional neural network (CNN) to incorporate features from ipsilateral and contralateral foot images. This CNN model achieves an area under the receiver operating characteristic curve (ROC-AUC) of 0.94 on the test set, demonstrating potential for early DFU detection. For RA, I use deep learning to automate the detection of joints and prediction of Sharp-van-der-Heijde (SvdH) scores on hand radiographs. I develop a two-stage system, with a joint detection model trained on pediatric hand radiographs from healthy individuals and a vision-transformer model trained on adult hand radiographs predicting erosion and joint-space-narrowing (JSN) scores. The system addresses class imbalance. The joint detection model achieves an F1 score of 0.98 on internal testing and ~0.92 mean precision on an external set. At best, the joint damage prediction model achieves balanced accuracy values of ~0.93 for erosion and JSN scores, and ~0.90 for individual wrist joints. In summary, my AI-based approaches show potential for early detection in DFU, and automated joint detection and scoring in RA. They have potential to improve detection, prediction and treatment of the DFU and RA.