In the medical field, AI development to support diagnosticians is advancing with an increase in the amount of medical image data such as CT, MRI, pathological image, endoscopic image, etc. Deep Learning can be cited as a core technology for computer aided diagnosis. Deep Learning is applied to various types of medical images （MRI, CT, chest X ray, ultrasonic, fundus, pathological, mammography, skin lesion examination, etc.）. In this paper, we introduce the findings of Deep Learning accumulated so far, centering on the latest research results, in the four categories of （1）tumor classification,（2） region extraction,（3） lesion detection, and（4）image quality improvement.

The detection of early signs of osteoporosis is required to maintain the quality of life of elderly people. Receiving osteoporosis screening by measurement of the bone mineral density （BMD） using dual-energy X-ray absorptiometry （DXA） is highly recommended, but the examination rates in Japan remain limited. In this study, we developed a method to estimate the BMD from chest radiographs and used it to diagnose osteoporosis. We randomly selected 63 men and 61 women who underwent chest radiography and DXA within two weeks and calculated the cortical thickness of the clavicle in radiographs. In 51 men and 49 women, we used multiple regression analysis to formulate to estimate the BMD of the lumbar vertebra and femur, actually measured by DXA based on the patient's body height and weight, body mass index, cortical thickness of the clavicle, and ratio between the cortical and cancellous thickness of the clavicle. Using these equations, we measured the BMD of the lumbar vertebra and femur in 12 men and 12 women and assessed the ability to diagnose osteoporosis in comparison with the diagnostic criteria. As a result, the accuracy of diagnosing osteoporosis was 66.7%. These results suggest that our method may be useful for osteoporosis screening.