AIP Conference Proceedings | 2018
The contribution describes progress in the long-term project concerning automatic diagnosis of spine bone lesions. There are two difficult problems: segmenting reliably possibly severely deformed vertebrae in the spine and then detect, segment and classify the lesions that are often hardly visible thus making even the medical expert decisions highly uncertain, with a large inter-expert variety. New approaches are described enabling to solve both problems with a success rate acceptable for clinical testing, at the same time speeding up the process substantially compared to the previous stage. The results are compared with previously published achievements.