Interaction between clinicians and artificial intelligence to detect fetal atrioventricular septal defects on ultrasound: how can we optimize collaborative performance?

OBJECTIVES: Artificial intelligence (AI) has shown promise in improving the performance of fetal ultrasound screening in detecting congenital heart disease (CHD). The effect of giving AI advice to human operators has not been studied in this context. Giving additional information about AI model workings, such as confidence scores for AI predictions, may be a way of further improving performance. Our aims were to investigate whether AI advice improved overall diagnostic accuracy (using a single CHD lesion as an exemplar), and to determine what, if any, additional information given to clinicians optimized the overall performance of the clinician-AI team. METHODS: An AI model was trained to classify a single fetal CHD lesion (atrioventricular septal defect (AVSD)), using a retrospective cohort of 121 130 cardiac four-chamber images extracted from 173 ultrasound scan videos (98 with normal hearts, 75 with AVSD); a ResNet50 model architecture was used. Temperature scaling of model prediction probability was performed on a validation set, and gradient-weighted class activation maps (grad-CAMs) produced. Ten clinicians (two consultant fetal cardiologists, three trainees in pediatric cardiology and five fetal cardiac sonographers) were recruited from a center of fetal cardiology to participate. Each participant was shown 2000 fetal four-chamber images in a random order (1000 normal and 1000 AVSD). The dataset comprised 500 images, each shown in four conditions: (1) image alone without AI output; (2) image with binary AI classification; (3) image with AI model confidence; and (4) image with grad-CAM image overlays. The clinicians were asked to classify each image as normal or AVSD. RESULTS: A total of 20 000 image classifications were recorded from 10 clinicians. The AI model alone achieved an accuracy of 0.798 (95% CI, 0.760-0.832), a sensitivity of 0.868 (95% CI, 0.834-0.902) and a specificity of 0.728 (95% CI, 0.702-0.754), and the clinicians without AI achieved an accuracy of 0.844 (95% CI, 0.834-0.854), a sensitivity of 0.827 (95% CI, 0.795-0.858) and a specificity of 0.861 (95% CI, 0.828-0.895). Showing a binary (normal or AVSD) AI model output resulted in significant improvement in accuracy to 0.865 (P < 0.001). This effect was seen in both experienced and less-experienced participants. Giving incorrect AI advice resulted in a significant deterioration in overall accuracy, from 0.761 to 0.693 (P < 0.001), which was driven by an increase in both Type-I and Type-II errors by the clinicians. This effect was worsened by showing model confidence (accuracy, 0.649; P < 0.001) or grad-CAM (accuracy, 0.644; P < 0.001). CONCLUSIONS: AI has the potential to improve performance when used in collaboration with clinicians, even if the model performance does not reach expert level. Giving additional information about model workings such as model confidence and class activation map image overlays did not improve overall performance, and actually worsened performance for images for which the AI model was incorrect. © 2024 The Authors. Ultrasound in Obstetrics & Gynecology published by John Wiley & Sons Ltd on behalf of International Society of Ultrasound in Obstetrics and Gynecology.

Grazing behaviour of dairy cows on biodiverse mountain pastures is more influenced by slope than cow breed.

The aim of this study was to determine how cows with different genetic merit behave and perform when grazing biodiverse and heterogeneous mountain pastures with different slopes. Three groups of 12 cows in late lactation, each composed of four Holstein, four Montbéliarde and four Valdostana Red Pied cows, breeds of increasing presumed robustness and decreasing milk yield (MY) potential. Cows grazed without concentrate either on a low-diversity flat pasture or on two species-rich mountainous pastures having slopes of either 7° or 22°. Milk yield, BW and grazing behaviour were monitored two times in the first and once in the second grazing cycle. Cows of different breeds had similar behaviour on all pastures. The Montbéliarde cows performed close to their production potential; Holstein and Valdostana cows produced less milk than anticipated. No breed difference in terms of BW loss was found. The Valdostana cows exhibited the least selective behaviour with respect to plant species and plant growth stage. Still, all cows searched for the most palatable vegetation regardless of pasture diversity. On the steep pasture, cows optimised the trade-off between ingesting and saving energy to obtain feed. They remained longer at the lowest zone and selected forbs, whereas cows on the flatter pasture went to the upper zone to select grasses. The present study gave no evidence for a superior short-term adaptation to harsh grazing conditions through an optimised feeding behaviour of the Valdostana breed compared to Montbéliarde and Holstein cows.

Frequent moving of grazing dairy cows to new paddocks increases the variability of milk fatty acid composition.

The aim of this work was to investigate the variations of milk fatty acid (FA) composition because of changing paddocks in two different rotational grazing systems. A total of nine Holstein and nine Montbéliarde cows were divided into two equivalent groups according to milk yield, fat and protein contents and calving date, and were allocated to the following two grazing systems: a long duration (LD; 17 days) of paddock utilisation on a heterogeneous pasture and a medium duration (MD) of paddock utilisation (7 to 10 days) on a more intensively managed pasture. The MD cows were supplemented with 4 kg of concentrate/cow per day. Grazing selection was characterised through direct observations and simulated bites, collected at the beginning and at the end of the utilisation of two subsequent MD paddocks, and at the same dates for the LD system. Individual milks were sampled the first 3 days and the last 2 days of grazing on each MD paddock, and simultaneously also for the LD system. Changes in milk FA composition at the beginning of each paddock utilisation were highly affected by the herbage characteristics. Abrupt changes in MD milk FA composition were observed 1 day after the cows were moved to a new paddock. The MD cows grazed by layers from the bottom layers of the previous paddock to the top layers of the subsequent new paddock, resulting in bites with high organic matter digestibility (OMD) value and CP content and a low fibre content at the beginning of each paddock utilisation. These changes could induce significant day-to-day variations of the milk FA composition. The milk fat proportions of 16:0, saturated FA and branched-chain FA decreased, whereas proportions of de novo-synthesised FA, 18:0, c9-18:1 and 18:2n-6 increased at paddock change. During LD plot utilisation, the heterogeneity of the vegetation allowed the cows to select vegetative patches with higher proportion of leaves, CP content, OMD value and the lowest fibre content. These small changes in CP, NDF and ADF contents of LD herbage and in OMD values, from the beginning to the end of the experiment, could minimally modify the ruminal ecosystem, production of precursors of de novo-synthesised FA and ruminal biohydrogenation, and could induce only small day-to-day variations in the milk FA composition.

Towards an agroecological assessment of dairy systems: proposal for a set of criteria suited to mountain farming.

Ruminant production systems have been facing the sustainability challenge, namely, how to maintain or even increase production while reducing their environmental footprint, and improving social acceptability. One currently discussed option is to encourage farmers to follow agroecological principles, that is, to take advantage of ecological processes to reduce inputs and farm wastes, while preserving natural resources, and using this diversity to increase system resilience. However, these principles need to be made more practical. Here, we present the procedure undertaken for the collaborative construction of an agroecological diagnostic grid for dairy systems with a focus on the mountain farming relying on the use of semi-natural grasslands. This diagnosis will necessarily rely on a multicriteria evaluation as agroecology is based on a series of complementary principles. It requires defining a set of criteria, based on practices to be recommended, that should be complied with to ensure agroecological production. We present how such agroecological criteria were identified and organized to form the architecture of an evaluation model. As a basis for this work, we used five agroecological principles already proposed for animal production systems. A group of five experts of mountain production systems and of their multicriteria evaluation was selected, with a second round of consultation with five additional experts. They first split up each principle into three to four generic sub-principles. For each principle, they listed three to eight categories of state variables on which the fulfilment of the principle should have a positive impact (e.g. main health disorders for the integrated health management principle). State variables are specific for a given production, for example, dairy farms. Crossing principles with state variables enabled experts to build five matrices, with 75 cells relevant for dairy systems. In each cell, criteria are specific to the local context, for example, mountain dairy systems in this study. Finally, we discuss the opportunities offered by our methodology, and the steps remaining for the construction of the evaluation model.

Milk fatty acid composition and cheese texture and appearance from cows fed hay or different grazing systems on upland pastures.

The objective of this work was to compare milk fatty acid (FA) profile and texture and appearance of Cantal cheeses obtained from cows grazing 2 different upland grasslands: a highly diversified pasture (74 species) of area 12.5 ha managed under continuous mode (C), and a weakly diversified pasture (31 species) of area 7.7 ha (an old temporary grassland) managed under rotational mode (R). A control group of cows fed a hay-based diet (indoors, I) was used. Three equivalent groups of 12 Montbéliarde cows underwent the 3 treatments from May to September 2008. The cheeses were manufactured during 3 consecutive days in early June, early July, and late August (27 cheeses in all). The texture, appearance, and chemical composition of the cheeses were determined after 12 wk of ripening. Concentrations of total saturated FA and monounsaturated FA were higher and lower, respectively, in I milks compared with pasture milks. The concentrations of trans-11-C18:1 and cis-9-C18:1, and polyunsaturated FA as well as yellowness decreased during the season in C-derived milk but remained constant in R-derived milk, through a combined effect of grass development stage and the cows' grazing selection. The I cheeses were, on average, firmer, less creamy, less elastic, and less yellow than the pasture cheeses. Decreasing and increasing trends in texture firmness during the season were observed for C and R cheeses, respectively. The rind of the pasture-fed cow cheese had fewer, less intensely colored, and less prominent spots than did that of I cheeses. This difference was probably due to greater migration of fat to the rind during pressing because of the lower fat melting point of the pasture-fed cow cheeses, which had higher unsaturated FA content. The greater amounts of fat deposited on the rind of the pasture-fed cow cheeses may have partially inhibited the microbial activity responsible for rind appearance. Our trial underlines the importance of the effects of grazing management associated with vegetation type on milk and cheese characteristics.

Effect of cattle grazing a species-rich mountain pasture under different stocking rates on the dynamics of diet selection and sward structure.

Although stocking rate is a key management variable influencing the structure and composition of pastures, only few studies have simultaneously analysed the seasonal patterns of pasture use by cattle, and the adjustments the animals make to maintain intake of a high-quality diet over the grazing season. Therefore, over a 3-year study, we recorded diet selection, plot use and impact of heifers on sward structure and quality under three different stocking rates (0.6, 1.0 and 1.4 livestock units (LU) per ha) in a species-rich mountain pasture of central France. Measurements were made on three occasions between early June and the end of September each year. Overall, heifers selected for bites dominated by legumes or forbs, and against reproductive grass, whatever the stocking rate or season. Selection for tall mixed (P < 0.05), short mixed (P < 0.05) and short pure grass bites (P < 0.01) was more pronounced in plots grazed at the lowest stocking rate. Although heifers' selection for short patches decreased at the end of the season (P < 0.001), they continued to graze previously grazed areas, thus exhibiting a typical 'patch grazing' pattern, with the animals that grazed at the lowest stocking rate tending to better maintain their selection for short patches in September (treatment × period: P = 0.078). Neither diet quality nor individual animal performance were affected by the different stocking rate treatments despite high variability in the quantity and quality of herbage offered and differences in diet selection. However, at the 1.4 LU per ha stocking rate, the quantity of forage available per animal at the end of the season, 0.79 t dry matter (DM) per ha of green leaves with the median of sward height at 4.6 cm, approached levels limiting cattle's ability to compensate for the effects of increasing stocking rate. In plots grazed at 0.6 LU per ha, the total herbage biomass remained higher than 3 t DM per ha with more than 30% of plot area still covered by reproductive grass patches at the end of the grazing season, which in the medium term should affect the botanical composition of these pastures. Sward heterogeneity was high in plots grazed at 1.0 LU per ha, with sufficient herbage availability (1.1 t DM per ha of green leaves) to maintain animal performance, and more than 15% of plot area was kept at a reproductive stage at the end of the grazing season. Hence, it could represent the optimal balance to satisfy both livestock production and conservation management objectives.

In vitro study on isolated tissues: applications to perinatology.

Experimentation with the "in vitro" methodology allowed the authors to study the effects of several pharmacological substances on human and animal tissue, making possible a qualitative assessment of activity, the quantitative dose-response relationships, the possible existence of specific drug receptors and the action, competitive or non-competitive, of a given antagonist. This methodology was applied to perinatology, particularly to two types of tissues: umbilical and placental vessels, arteries and veins; human and animal myometrium. Among the known circulating vasoactive substances, the authors obtained helpful results with the use of morphine, with obvious implications in obstetrical practice, that opioid substances very frequently are used to control the pains of childbirth. The experimentation with alcohol on isolated umbilical arteries and veins helps to explain the fetal alcohol syndrome. Finally, the study "in vitro" on human and animal myometrial strips gave interesting cues for the research into the regulation of the uterine contraction.