Artificial Intelligence in Veterinary Imaging: An Overview.

Artificial intelligence and machine learning have been increasingly used in the medical imaging field in the past few years. The evaluation of medical images is very subjective and complex, and therefore the application of artificial intelligence and deep learning methods to automatize the analysis process would be very beneficial. A lot of researchers have been applying these methods to image analysis diagnosis, developing software capable of assisting veterinary doctors or radiologists in their daily practice. This article details the main methodologies used to develop software applications on machine learning and how veterinarians with an interest in this field can benefit from such methodologies. The main goal of this study is to offer veterinary professionals a simple guide to enable them to understand the basics of artificial intelligence and machine learning and the concepts such as deep learning, convolutional neural networks, transfer learning, and the performance evaluation method. The language is adapted for medical technicians, and the work already published in this field is reviewed for application in the imaging diagnosis of different animal body systems: musculoskeletal, thoracic, nervous, and abdominal.

Femoral parallelism: evaluation and impact of variation on canine hip dysplasia assessment.

Adequate radiographic positioning on the X-ray table is paramount for canine hip dysplasia (HD) screening. The aims of this study were to evaluate femoral parallelism on normal ventrodorsal hip extended (VDHE) view and the effect of femoral angulation (FA) on Norberg Angle (NA) and Hip Congruency Index (HCI). The femoral parallelism was evaluated comparing the alignment of the long femoral axis with the long body axis in normal VDHE views and the effect of FA on NA and HCI on repeated VDHE views with different levels of FA. The femoral long axis in normal VDHE views showed a ranged of FA from -4.85° to 5.85°, mean ± standard deviation (SD) of -0.06 ± 2.41°, 95% CI [-4.88, 4.76°]. In the paired views, the mean ± SD femur adduction of 3.69 ± 1.96° led to a statistically significant decrease NA, and HCI, and femur abduction of 2.89 ± 2.12 led to a statistically significant increase in NA and HCI (p < 0.05). The FA differences were also significantly correlated with both NA differences (r = 0.83) and HCI differences (r = 0.44) (p < 0.001). This work describes a methodology that allows evaluation of femoral parallelism in VDHE views and the results suggest that femur abduction yielded more desirable NA and HCI values and adduction impaired NA and HCI values. The positive linear association of FA with NA and HCI allows the use of regression equations to create corrections, to reduce the influence of poor femoral parallelism in the HD scoring.

Adaptive Feature Recombination and Recalibration for Semantic Segmentation With Fully Convolutional Networks.

Fully convolutional networks have been achieving remarkable results in image semantic segmentation, while being efficient. Such efficiency results from the capability of segmenting several voxels in a single forward pass. So, there is a direct spatial correspondence between a unit in a feature map and the voxel in the same location. In a convolutional layer, the kernel spans over all channels and extracts information from them. We observe that linear recombination of feature maps by increasing the number of channels followed by compression may enhance their discriminative power. Moreover, not all feature maps have the same relevance for the classes being predicted. In order to learn the inter-channel relationships and recalibrate the channels to suppress the less relevant ones, squeeze and excitation blocks were proposed in the context of image classification with convolutional neural networks. However, this is not well adapted for segmentation with fully convolutional networks since they segment several objects simultaneously, hence a feature map may contain relevant information only in some locations. In this paper, we propose recombination of features and a spatially adaptive recalibration block that is adapted for semantic segmentation with fully convolutional networks- the SegSE block. Feature maps are recalibrated by considering the cross-channel information together with spatial relevance. The experimental results indicate that recombination and recalibration improve the results of a competitive baseline, and generalize across three different problems: brain tumor segmentation, stroke penumbra estimation, and ischemic stroke lesion outcome prediction. The obtained results are competitive or outperform the state of the art in the three applications.