Radiology is central to modern diagnosis, yet growing scan volumes and uneven access to specialists strain its capacity, particularly in India. This paper examines how artificial intelligence (AI), mainly deep learning, is being used to improve diagnostic accuracy and to detect disease earlier. Drawing on randomised, prospective and systematic evidence published largely between 2023 and 2025, it considers three areas: breast cancer screening, chest imaging for lung cancer, and tuberculosis screening. Large studies in Sweden and Germany indicate that AI-supported mammography reading can raise cancer detection while reducing radiologist workload and without inflating recall. Reviews of lung nodule detection report high sensitivity, although much of the evidence is retrospective. Indian implementation work suggests AI can add measurable tuberculosis yield and appears cost-effective, but deployment is hampered by infrastructure and workflow problems. Key barriers include automation bias, dataset shift, limited external validation and unclear accountability. The paper concludes that AI performs best as a supervised second reader within monitored workflows, and that prospective, locally validated trials should be the priority for responsible adoption.