Automated classification of protein crystallization images using support vector machines with scale-invariant texture and Gabor features.

Protein crystallography laboratories are performing an increasing number of experiments to obtain crystals of good diffraction quality. Better automation has enabled researchers to prepare and run more experiments in a shorter time. However, the problem of identifying which experiments are successful remains difficult. In fact, most of this work is still performed manually by humans. Automating this task is therefore an important goal. As part of a project to develop a new and automated high-throughput capillary-based protein crystallography instrument, a new image-classification subsystem has been developed to greatly reduce the number of images that require human viewing. This system must have low rates of false negatives (missed crystals), possibly at the cost of raising the number of false positives. The image-classification system employs a support vector machine (SVM) learning algorithm to classify the blocks making up each image. A new algorithm to find the area within the image that contains the drop is employed. The SVM uses numerical features, based on texture and the Gabor wavelet decomposition, that are calculated for each block. If a block within an image is classified as containing a crystal, then the entire image is classified as containing a crystal. In a study of 375 images, 87 of which contained crystals, a false-negative rate of less than 4% with a false-positive rate of about 40% was consistently achieved.