AI platform for pixel spacing, distance, and volumetric predictions from dental images
A machine learning model is trained to predict pixel spacing, distance, and volumetric measurements. Training images are obtained by inpainting around an original image and scaling the inpainted image to obtain the training image having a different pixel spacing than the original image. The machine learning model may include an encoder, a transformer, a first TC layer, and a second TC layer. During training, loss may be obtained from a comparison of the output to the first TC layer to a coarse pixel spacing matrix and a comparison of the output of the second TC layer to a fine pixel spacing matrix. During utilization, the pixel spacing of an image may be obtained using the machine learning model and used to correct the image or measurements obtained from the image.
1. A method for training a machine learning model comprising:
providing a plurality of training data entries, each including a training image and a target pixel spacing matrix, the target pixel spacing matrix corresponding to real-world distances represented by pixels of the training image;
for each training data entry of the plurality of training data entries:
processing the training image of each training data entry using the machine learning model to obtain an estimated pixel spacing matrix; and
updating the machine learning model according to differences between the target pixel spacing matrix and the estimated pixel spacing matrix.
2. The method of claim 1 , wherein:
the target pixel spacing matrix is a coarse target pixel spacing matrix;
each training data entry of the plurality of training data entries further includes a fine target pixel spacing matrix that is larger than the coarse target pixel spacing matrix, the fine target pixel spacing matrix corresponding to the real-world distances represented by the pixels of the training image and having a higher resolution than the coarse target pixel spacing matrix;
the machine learning model includes a first model portion and a second model portion;
the method further comprises:
processing the training image of each training data entry with the first model portion to obtain the estimated pixel spacing matrix;
processing the estimated pixel spacing matrix with the second model portion to obtain a fine estimated pixel spacing matrix; and
updating the first model portion and the second model portion according to the differences between the estimated pixel spacing matrix and the coarse target pixel spacing matrix and differences between the fine estimated pixel spacing matrix and the fine target pixel spacing matrix.
3. The method of claim 2 , wherein the first model portion includes an encoder followed by a first transpose convolution stage.
4. The method of claim 3 , wherein the second model portion is a second transpose convolution stage.
5. The method of claim 1 , wherein the training image of each training data entry is obtained by a modality selected from the group consisting of full mouth series X-rays (FMX), dental cone beam computed tomography (CBCT), cephalometric X-ray, intra-oral optical image, panoramic dental X-ray, dental magnetic resonance imaging (MRI) image, dental light detection and ranging (LIDAR) image.