Our database consists of CT data shared with us from Harvard University / Massachusetts General Hospital.
The images are of cadavers, acquired at various voltage settings, noise levels, and slice thicknesses, and reconstructed using representative methods.
These images can be used for machine learning and education. For example, there are several versions of images of the same cadaver with different noise levels for development of denoising methods.
The images are organized into folders by body part: head, chest, and abdomen. In the folders, files are organized by cadaver ID and then by the reconstruction algorithm, voltage, noise, and slice thickness. The file name 100K10NI2_5MMSS00_0351 indicates the voltage (100K), noise index (10NI), slice thickness(2_5MM). Also, SS00 indicates the filtered backprojection method, SS50 denotes the statistical iterative method, and VEO the Veo model-based reconstruction method.
Our team members at RPI and MGH indeed intended to share these images publicly with the community. However, we have been recently advised by MGH IRB with several considerations: mainly, (1) the deceased families were not consented for that level of sharing; (2) the head and neck information can be reconstructed back to give the face of the person scanned; etc. Given how much is already available on NIH public data sites and AAPM data sites, we suggest those who are interested in our database to seek alterantives. We are very sorry for any inconvenience.