Feasibility of the use of deep learning classification of teat-end condition in Holstein cattle.

Infections with pathogenic bacteria entering the mammary gland through the teat canal are the most common cause of mastitis in dairy cows; therefore, sustaining the integrity of the teat canal and its adjacent tissues is critical to resist infection. The ability to monitor teat tissue condition is a key prerequisite for udder health management in dairy cows. However, to date, routine assessment of teat condition is limited to cow-side visual inspection, making the evaluation a time-consuming and expensive process. Here, we demonstrate a digital teat-end condition assessment by way of deep learning. A total of 398 digital images from dairy cows' udders were collected on 2 commercial farms using a digital camera. The degree of teat-end hyperkeratosis was scored using a 4-point scale. A deep learning network from a transfer learning approach (GoogLeNet; Google Inc., Mountain View, CA) was developed to predict the teat-end condition from the digital images. Teat-end images were split into training (70%) and validation (15%) data sets to develop the network, and then evaluated on the remaining test (15%) data set. The areas under the receiver operator characteristic curves on the test data set for classification scores of normal, smooth, rough, and very rough were 0.778 (0.716-0.833), 0.542 (0.459-0.608), 0.863 (0.788-0.906), and 0.920 (0.803-0.986), respectively. We found that image-based teat-end scoring by way of deep learning is possible and, coupled with improvements in image acquisition and processing, this method can be used to assess teat-end condition in a systematic and efficient manner.

Technical note: A digital technique and platform for assessing dairy cow teat-end condition.

Because infections with pathogenic bacteria entering the mammary gland through the teat canal are the most common cause of mastitis in dairy cows, sustaining the integrity of the teat canal and its adjacent tissues is critical to resist infection. The ability to monitor teat tissue condition is therefore a key prerequisite for udder health management in dairy cows. However, to date, routine assessment of teat-end condition is limited to cow-side visual inspection, making the evaluation a time-consuming and expensive process. Here, we illustrate and demonstrate a method for assessing teat-end condition of dairy cows through digital images and software. A digital workflow has been designed where images of dairy cow teats are obtained and processed to display individual teats, and the cow and teat images are labeled and displayed through a graphical user interface. The interface then allows an evaluator to assess quarter- and cow-level teat-end condition and store the results for review and future analysis. The digital workflow permits several advantages such as the ability to perform remote teat-end condition assessments, and assess inter- and intrarater variability of teat-end condition scoring. We demonstrate the image-based teat-end condition assessment of 194 dairy cows that also had cow-side teat-end condition assessments by 2 expert evaluators. Weighted Cohen's kappa statistic (κ) was computed to measure the evaluators' concordance of categorical scores of quarter- and cow-level assessments when using cow-side and image-based assessments. Substantial agreement (0.61 ≤ κ ≤ 0.80) was observed between an evaluator's cow-side and image-based assessments at the quarter and cow level. Moderate agreement (0.41 ≤ κ ≤ 0.60) was observed between evaluators when using image-based assessments at the quarter and cow level. Near perfect agreement (κ = 0.89, 95% confidence interval 0.78-1.00) was observed between evaluators when using cow-side assessments at the quarter level, and substantial agreement (κ = 0.66, 95% confidence interval 0.53-0.79) was observed when using cow-side assessments at the cow level. This suggests that image-based teat-end condition classification is possible, and coupled with improvements in image acquisition and image processing, this method can be used to assess teat-end condition in a systematic and convenient manner.

Using infrared thermography to determine changes in teat skin surface temperature after machine milking in dairy cows.

Mastitis is among the costliest diseases affecting dairy cows, partly due to the resulting permanent reduction in the quantity and quality of milk produced. Most mastitis cases involve pathogenic organisms entering the cow's mammary gland through the teat canal. The teat has natural defenses against these pathogens that can be disrupted during milk harvesting. These disruptions of the teat's circulatory system and tissue integrity can predispose them to mastitis. Traditionally, machine milking-induced changes in teat blood circulation and tissue integrity have been assessed by means of manual evaluation and ultrasonography. Infrared thermography has previously been shown to produce precise and consistent measurements of skin surface temperatures (SST) on cows' hind teats. Our objective was to describe the variability in the teat SST following machine milking. Describing the variability in teat SST before and after milking could be useful to guide further studies to elucidate the physiology of the effects of milking on teat defense mechanisms. In this observational study, thermographic images of both hind teats from 140 cows immediately pre- and postmachine milking were analyzed. The average SST were subsequently determined at the proximal, middle, and distal aspects of each hind teat using image analysis software. The LSM (95% CI) from general linear mixed models of the pre- and postmilking SST, respectively, were 33.6 (33.5-33.8)°C and 35.4 (35.3-35.5)°C at the proximal aspect, 33.2 (33.1-33.4)°C and 35.2 (35.1-35.3)°C at the middle aspect, and 32.3 (32.1-32.5)°C and 34.0 (33.9-34.1)°C at the distal aspect. The observed increase in SST from pre- to postmilking SST at all 3 aspects of the teat suggest that some of the variability in the SST can be attributed to the milking event. Future research is warranted to investigate the biological relevance of SST changes during machine milking and any potential change in teat defense mechanisms, risk of mastitis, or other pathologies.

Development and evaluation of a standardized technique to assess teat skin temperature of dairy cows using infrared thermography.

Much research has been done to develop methods to assess dimensional teat traits in dairy cows. In contrast, diagnostic techniques to reliably assess the circulatory system of teats are limited. Infrared thermography facilitates measurements of skin temperature and could be used to detect physiological and pathological changes to the teat tissue associated with machine milking, as temperature reflects the underlying blood circulation and tissue metabolism. Our objective was to develop and evaluate a scanning technique to quantify teat skin temperature in dairy cows using infrared thermography. Using a portable thermography camera, 2 operators obtained duplicate scans of both hind teats from 20 Holstein cows, resulting in 80 thermographic images (20 cows × 2 operators × 2 images). Average teat skin temperatures at the proximal, middle, and distal teat aspects were determined. We used Pearson correlation coefficients (r), intraclass correlation coefficients (ICC), and concordance correlation coefficients (CCC) to assess interoperator reproducibility (i.e., agreement between measurements performed by different operators) and intraoperator repeatability (i.e., agreement between measurements performed by the same operator). Pearson correlation coefficients revealed a very strong correlation for measurements at the proximal, middle, and distal aspects of the teat, respectively, between operators (r ≥0.95) and duplicate scans (r ≥0.94) within operators. Intraclass correlation coefficients and CCC indicated excellent interoperator reproducibility (ICC ≥0.95, CCC ≥0.95) and excellent intraoperator repeatability (ICC ≥0.94, CCC ≥0.94), respectively, for measurements at all 3 aspects. Least squares means (95% confidence interval) for average teat skin temperatures at the proximal, middle, and distal teat aspects, respectively, were 33.2 (32.6-33.8), 32.4 (31.5-33.2), and 30.9 (29.8-32.0) °C for operator 1, and 33.2 (32.6-33.8), 32.4 (31.6-33.3), and 31.0 (29.9-32.0) °C for operator 2. Average temperatures between duplicate scans within operators at the proximal, middle, and distal aspects, respectively, were 33.3 (32.7-33.9), 32.5 (31.7-33.3), and 31.0 (29.9-32.1) °C for the first scan and 33.2 (32.6-33.8), 32.3 (31.5-33.1), and 30.8 (29.7-31.9) °C for the second scan. We conclude that infrared thermography facilitates precise measurements of skin temperatures of cows' hind teats.

Delta thermal radiomics: An application in dairy cow teats.

We describe a novel approach for analyzing thermal images by way of radiomics (i.e., thermal radiomics) and how it can be used to monitor short-term temperature changes of dairy cow hind teats; that is, delta thermal radiomics. The heat generated from metabolic activities and blood-flow patterns can be visualized using thermal radiography of the skin surface. The hind teats from 25 dairy cows were imaged with a digital thermal camera and the images were converted to medical images (DICOM format) by mapping the multi-channel colorized thermal image to a monochromatic image whose intensities represent temperature. The 50 teats (left and right hind) were then manually segmented by 2 investigators. Radiomics analysis, which is a common method of extracting semantic and nonsemantic image biomarkers from medical images for machine learning, was performed. To evaluate whether this approach can detect pre- and postmilking differences, 18 cows were imaged before and after milking, the teats were manually segmented, and radiomic calculations were performed. Student's t-test was used to provide an estimate of the likelihood of whether postmilking thermal image biomarkers are the same as premilking thermal image biomarkers, and Cohen's d was used to evaluate the size of the effect (d > 1.2). To evaluate uncertainties from manual segmentation, the Dice similarity score (DS) between the 2 investigators' segments was computed. The average DS (95% confidence limit) was 0.952 (0.913-0.982) when comparing the 2 investigators' segmentations. There was no significant difference in DS when comparing the left and right segmented teats, suggesting that teats can be segmented consistently. No differences (d < 0.36) were observed when comparing image biomarkers from one investigator's segments with the other's, suggesting that image biomarkers computed from one investigator's segmentation of teats are not likely to differ from those computed from the other investigator. When comparing image biomarkers before and after milking, 109 image biomarkers were analyzed, and 17 image biomarkers were simultaneously significant and exhibited effect size. Thus, delta thermal radiomics offers a noninvasive and quantitative method of monitoring skin temperature changes in humans and animals after an intervention. The advantage of this approach is that it can reveal both perceptible and imperceptible surface temperature features that may be useful for detecting and managing dairy teat health.

Phase I/II study of a five-fraction hypofractionated accelerated radiotherapy treatment for low-risk localised prostate cancer: early results of pHART3.

AIMS: Most men with low-risk localised prostate cancer prefer treatments with high control rates and minimal disruption to their lives. Hypofractionating external radiation treatments can theoretically maintain high bioequivalent tumour doses, decrease treatment visits and decrease acute and late toxicities. The aim of this study was to assess the toxicity and feasibility of a hypofractionated accelerated regimen for these patients. MATERIALS AND METHODS: The present study was a phase I/II study in which patients with T1-2b, Gleason < or = 6 and prostate-specific antigen (PSA) < or = 10 ng/ml prostate cancer received 35Gy in five fractions, once a week over 29 days. Treatment was delivered with intensity-modulated radiotherapy on standard linear accelerators, with daily image guidance using gold seed fiducials, and a 4mm clinical target volume to planning target volume margin. RESULTS: As of January 2008, the target accrual of 30 patients had been reached and all had completed treatment and at least 6 months of follow-up. Dose-volume histogram objectives were achievable in all patients. Treatment was very well tolerated with no grade 3 or 4 genitourinary toxicity, gastrointestinal toxicity nor fatigue observed (95% confidence interval 0-12%). As a group, compared with baseline, the following additional grade 2 toxicities were observed: 13% genitourinary, 7% gastrointestinal and 10% fatigue. At 6 months all scores had returned to or improved over baseline. The median PSA before treatment was 6.0 ng/ml. At 6 months, the median PSA was 1.8 ng/ml and 75% had a PSA < or = 3.0 ng/ml. CONCLUSIONS: This novel technique using standard linear accelerators seems feasible and is well tolerated. Further follow-up will be carried out to document late toxicity and efficacy.

Evaluation of optimized compensators on a 3D planning system.

A commercially available treatment planning system contains several functions that allow for the automation of missing tissue and optimized compensators, where the former retracts the bolus toward the source, and the latter attempts, by iteration, to establish a uniform dose at some user defined depth. The intent of this paper is to report on the compensators designed by the system and to compare them to those devised through conventional techniques. It is demonstrated that the system can model the dosimetric effects of compensators with a high degree of accuracy; measured and predicted doses agree to within 3%. Optimized compensators show slightly improved dose uniformity over thickness reduced compensators. Both show significantly improved uniformity over compensators that simply retract the bolus geometry. In cases where internal inhomogeneities exist, however, the dose uniformity from the optimized compensators vary by as much as 6% at the target depth. These deviations are comparable to the errors of the inhomogeneity algorithm itself. The pathlength reduction technique has been applied to both missing tissue and inhomogeneity compensation, and it has been found that for inhomogeneity compensation, the pathlength reduced compensators produce more uniform distributions than those generated by the optimization algorithm.

Comparison of helical and average computed tomography for stereotactic body radiation treatment planning and normal tissue contouring in lung cancer.

AIMS: To compare average computed tomography (CT(AVE)) datasets with free breathing helical computed tomography (CT(HEL)) for contouring organs at risk (OARs) and radiation treatment planning in patients receiving stereotactic body radiation therapy (SBRT) in the lung. MATERIALS AND METHODS: Ten SBRT patients with peripheral stage I non-small cell lung cancer underwent a CT(HEL) and a four-dimensional computed tomography scan in the treatment position. CT(AVE) datasets were generated from the four-dimensional computed tomography scan. The following OARs were delineated on the CT(HEL) and CT(AVE) datasets of each patient: lung minus internal target volume, trachea/main bronchus, heart, oesophagus and spinal cord. Volumes and geometric centres of the OARs, as well as the dosimetric impact of planning with these different datasets, were examined. RESULTS: There were no statistical differences in the OAR geometric centre coordinates nor in the OAR volumes between the CT(HEL) and CT(AVE) datasets, except that CT(AVE)-defined trachea/main bronchus and lung minus internal target volume mean volumes were larger than those defined on the CT(HEL) (46, 43 cm³ and 3516, 3378 cm³, respectively, P<0.05). Despite this, there were no significant differences in the mean and maximum doses to the OAR contours when using the CT(HEL) or CT(AVE) for planning (<4% average change in the maximum and mean doses for all OARs, P>0.05). There were also no significant differences in the locations of the hotspots within OARs among the CT(HEL) or CT(AVE) datasets (P>0.05). CONCLUSIONS: CT(AVE) datasets may be used in place of CT(HEL) for OAR contouring and dose calculations. When four-dimensional computed tomography is available, it may not be necessary to acquire a separate CT(HEL) scan for OAR contouring or dosimetric purposes.