Improved Brain Segmentation Using Pixel Separation and Additional Segmentation Features.

Brain segmentation is key to brain structure evaluation for disease diagnosis and treatment. Much research has been invested to study brain segmentation. However, prior research has not considered separating actual brain pixels from the background of brain images. Not performing such separation may (a) distort brain segmentation models and (b) introduce overhead to the modeling performance. In this paper, we improve the performance of brain segmentation using 3D, fully Convolutional Neural Network (CNN) models. We use (i) infant and adult datasets, (ii) a multi-instance loss method to separate actual brain pixels from the background and (iii) Gabor filter banks and K-means clustering to provide additional segmentation features. Our model obtains dice coefficients of \(87.4\%\)–\(94.1\%\) (i.e., an improvement of up to 11% to the results of five state-of-the-art models). Unlike prior studies, we consult experts in medical imaging to evaluate our segmentation results. We observe that our results are fairly close to the manual reference. Moreover, we observe that our model is 1.2x–2.6x faster than prior models. We conclude that our model is more efficient and accurate in practice for both infant and adult brain segmentation.