Search

The need for huge data sets represents a bottleneck for the application of artificial intelligence. Substantially fewer annotated target lesions than normal tissues for comparison present an additional problem in the field of pathology. Organic brains overcome these limitations by utilizing large numbers of specialized neural nets arranged in both linear and parallel fashion, with each solving a restricted classification problem. They rely on local Hebbian error corrections as compared to the nonlocal back-propagation used in most artificial neural nets, and leverage reinforcement. For these reasons, even toddlers are able to classify objects after only a few examples. Rather than provide an overview of current AI research in pathology, this review focuses on general strategies for overcoming the data bottleneck. These include transfer learning, zero-shot learning, Siamese networks, one-class models, generative networks, and reinforcement learning. Neither an extensive mathematic background nor advanced programing skills are needed to make these subjects accessible to pathologists. However, some familiarity with the basic principles of deep learning, briefly reviewed here, is expected to be useful in understanding both the current limitations of machine learning and determining ways to address them.

The use of abrasive waterjets (AWJ) to machine metal surfaces is prevalent through-out industry. AWJ requires the purchase and disposal of costly abrasive particles which increases manufacturing costs and environmental-footprint. This motivates the exploration for low-cost, eco-friendly alternatives. A novel method which utilizes a stream of high-speed water droplets (WD) travelling through a sub-atmospheric pressure environment to remove material without using abrasive grit is presented. In this paper, the efficacy of this new process is investigated and compared to AWJ and pure water jet (WJ) (i.e., without abrasive particles) by measuring the material removal rates (MRR) of each process on a hard-to-machine steel, DP 1180, for equivalent water-flow conditions. Trench profiles are created by traversing each of the jets across a metal sample, from which the MRR and trench profiles are obtained. The MRR results reveal that WD is significantly faster than WJ, but is subordinate compared to AWJ for MRR efficiency. This identifies a trade-off between manufacturing cost and time.

Integrated photoelasticity is investigated for a soft material subjected to a three-dimensional stress state. In the experiment, a solid sphere is pressed against a gelatin gel (Young's modulus is about 4.2 kPa) that deforms up to 4.5 mm depending on the loading forces. The resulting photoelastic parameters (phase retardation, azimuthal angle, and stress-optic coefficient) in the gel are measured using a polarization camera. The measured retardation and azimuth are compared with the analytical prediction based on Hertzian contact theory. Remarkably, experimental and analytical results of the photoelastic parameters show a reasonable agreement not only in the retardation but also in the azimuth that is related to the direction of principal stresses and but rarely validated in previous studies, is essential for reconstructing three-dimensional stress fields in soft materials. The stress-optic coefficient of the gelatin gel used is 3.12$\times10^{-8}$ 1/Pa. Such findings proved that integrated photoelasticity is useful for measuring the three-dimensional stress field in soft materials, which is of importance in biomedical engineering and cell printing applications., Comment: 15 pages, 13 figures

Abrasive waterjet (AWJ) cutting is a manufacturing technique, which uses a high-speed waterjet as the transport medium for abrasive particles to erode and cut through metal workpieces. The use of abrasives has significant environmental impacts and leads to the high operating costs of AWJ cutting. Therefore, it is important to investigate whether other metal cutting approaches can perform the same tasks with reduced environmental and economic impacts. One such manufacturing innovation is water droplet machining (WDM). In this process, the waterjet, which is immersed in a sub-atmospheric pressure environment, is discretized into a train of high velocity water droplets, which are able to erode and cut through the metal workpiece without abrasives. However, the cutting velocity of WDM is two orders of magnitude slower than AWJ. In this paper, a comparative life cycle and life cycle cost assessments were performed to determine which waterjet cutting technology is more beneficial to the environment and cost-efficient, considering their impacts from cradle to grave. The results show lower environmental and economic impacts for AWJ compared to WDM due to the AWJ’s ability to cut more metal over the service life than the WDM. Further sensitivity analyses give insight into how the change in abrasive rate is the most sensitive input for the AWJ, whereas the machine lifetime and electricity usage are the most sensitive inputs for the WDM. These results provide a valuable comparison between these alternative waterjet cutting technologies.

The transient force exerted by a low-speed liquid droplet impinging onto a flat rigid surface is investigated experimentally. The measurements employ a high-sensitivity piezo-electric sensor, along with a high-speed camera, and cover four decades in droplet Reynolds number and greater than two decades in Weber number. Across these ranges, the peak of individual force profiles span from 3 mN to over 1300 mN. Once normalised, the force–time profiles support the existence of an inertially dominated self-similar regime. Within this regime, previous numerical and theoretical studies predict a $\sqrt{t}$ dependence of impact normal force during the initial pre-peak rise. While our measurements confirm this finding, they also indicate that, after the peak force the profiles exhibit an exponential decay. This long-time decay law suggests treatment of the momentum transport from the droplet using a lumped model. An observed linear dependence between the force and force decay rate supports this approach. The reason for the efficacy of treating this system via a lumped model apparently connects to the physics right at the surface that limit the rate of momentum transport from the droplet to the surface. This is explored by estimating the momentum transfer by solely using the deforming droplet shape, but under the condition of negligible momentum gradients within the droplet. The short- and long-time solutions are combined and the resulting model equation is shown to accurately cover the entire force–time profile.

Though initially outclassed by more computationally efficient machine learning algorithms, neural network–based deep learning strategies have come of age with modern implementations and hardware and are now capable of addressing a range of previously intractable problems. These include methods for dealing with complex spatial or sequential data, strategies for coping with sets of data in which not all of the examples are annotated, and ways of generalizing across different databases. Other methods take inspiration from biological strategies such as operant conditioning or evolutionary principles. These and other techniques described here form the basis for many of the advances in the use of artificial intelligence in the service of pathology, as well as medicine in general.

In this work, we present a machine learning method to guide an ultrasound operator towards a selected area of interest. Unlike other automatic medical imaging methods, ultrasound imaging is one of the few imaging modalities where the operator’s skill and training are critical in obtaining high quality images. Additionally, due to recent advances in affordability and portability of ultrasound technology, its utilization by non-experts has increased. Thus, there is a growing need for intelligent systems that have the ability to assist ultrasound operators in both clinical and non-clinical scenarios. We propose a system that leverages machine learning to map real time ultrasound scans to transformation vectors that can guide a user to a target organ or anatomical structure. We present a unique training system that passively collects supervised training data from an expert sonographer and uses this data to train a deep regression network. Our results show that we are able to recognize anatomical structure through the use of ultrasound imaging and give the user guidance toward obtaining an ideal image.

The plastic flow and anisotropy of a low-carbon steel (AISI 1018) across a range of strain-rates is probed using compression experiments. Material models are then calibrated from these results. The strain-rates examined were 10−3 /s, 1 /s and 3.5 × 103 /s. It was found that the strain-hardening behavior is similar between the 10−3 /s and 1 /s families of experiments, with the latter showing a higher flow stress, as expected. However, the 3.5 × 103 /s experiments show not only an even higher flow stress, but a different strain-hardening behavior as well: they exhibit almost perfect plasticity, due to deformation-induced thermal-softening at larger strains. Comparing experiments in 3 mutually perpendicular directions, some mild plastic anisotropy was discovered. The initial yield and plastic flow of the material in 2 of these directions is comparable, and higher than in the 3rd one. It was also found that the plastic anisotropy evolves differently between the three directions, and also non-linearly with the strain-rate. These results were used to calibrate 3 versions of the Johnson–Cook (J–C) material model (original and 2 modified). The J–C modifications involve strain-rate-dependent model coefficients and exponents, as well as modified functional dependencies of the flow stress on the strain-rate and yield the best agreement with the experiments across the 6 orders of magnitude variation of the strain-rate. Assuming symmetry in tension and compression, and pressure-independence of yield, the plastic anisotropy was represented using the Hill 1948 yield criterion and 2 versions of the Karafillis–Boyce (K–B) one. The calibrations were performed at 5 levels of reference true strain: 0.075, 0.1, 0.15, 0.2, 0.25. For the material and experiments in this work, the Hill criterion provides good representation of anisotropy in the stress space. To capture the evolving plastic anisotropy, the parameters of the Hill criterion were evolved with plastic strain. Despite this, the criterion failed to provide accurate descriptions of the shape of the deformed specimens. In contrast, the performance of the 2 versions of K–B is significantly better, which would justify the higher effort required for calibration.

Impacting water droplets are capable of eroding steam turbine blades, high speed aircraft and even serve as a machining operation. In this latter process, high velocity (∼100m/s) water droplets impact a solid surface under conditions in which the ambient air pressure is sub-atmospheric. The physics of this erosion mechanism is not completely understood. To begin to unravel these physics, we present results pertaining to the effects of ambient air pressure on the impact force of low velocity (1m/s) droplets. It is well known that droplet splash is suppressed when the ambient air pressure is reduced. The effects on the associated impact force are, however, unknown. In this article we examine the impact force of 3.5 mm diameter, low velocity droplets in a reduced pressure environment. A 38x38x50 cm vacuum chamber fitted with a unidirectional piezoelectric force sensor to measure the transient force of impacting droplets. Preliminary results in atmospheric air pressure reveal that the impulse of the droplets depends linearly on impact velocity while the average impact force depends quadratically on impact velocity. The results under reduced pressures are compared and discussed relative to the underlying physics.

The surface-normal impingement of a liquid jet emanating from a nozzle is of use in numerous technological applications (e.g., surface cleaning, waterjet cutting, and needle-free injection). Upon impact, the (Rayleigh) jet is characterized as either a steady-state (time-invariant) jet, a wavy jet, or a droplet train. The present study experimentally investigates the force imparted by these three distinctly different processes for identical nozzle flow conditions (i.e., velocity and jet diameter). Force models are developed based on Rayleigh jet theory, and they are shown to compare well with the measurements. The force induced by the wavy jet is sinusoidal, oscillating about the steady-state force, while the droplet train force is discrete with periods of zero induced force. Due to conservation of momentum, the peak force experienced by the droplet train is significantly higher than that of the steady stream. This identifies the droplet train as the dominant process to obtain higher peak forces than the continuous stream counterparts during impact with a surface. This can then be used as a means to increase the performance of applications such as waterjet cutting and surface cleaning.

The surface normal impingement of a cylindrical liquid jet emanating from a nozzle is of use in numerous technological applications (e.g.., waterjet cutting). If a greater distance between the nozzle and the impingement surface is allowed, then the initially continuous liquid jet may fragment into a train of discrete droplets, whose peak impact force can exceed that of an unbroken continuous liquid jet. The present study experimentally investigates the force imparted by these two distinctly different processes for identical flow conditions (i.e., velocity, jet diameter, etc.). Based on the conservation of momentum, a justification for the significantly higher peak forces observed in the droplet train case is presented.

Impacting water droplets are capable of eroding soil, rock, turbine blades and even high speed aircraft. Research has shown that high velocity water droplet impingement onto a solid workpiece can strip paint, remove rust, and serve as a machining operation. This technique is different from waterjet cutting as a train of water droplets are used to transport momentum to a workpiece rather than a continuous jet. Also, no abrasive medium is used which produces an environmentally friendly process. The exploitation of this water droplet impact phenomenon in industrial applications as a means to deform and remove material does not currently exist. If the impact dynamics of water droplets can be understood and controlled, then industries would have the framework upon which they could employ this phenomenon in novel manufacturing equipment. In this paper, as a starting point, the impact force of 2.9mm diameter water droplets impacting at low velocities (∼2m/s) was studied experimentally and numerically. A piezoelectric force sensor was used to measure the transient force of impacting water droplets, while numerical simulations were used to identify the solid-fluid interaction and develop the basis for high velocity (>100m/s) liquid droplet impact dynamics. Agreement was established between experimental observations and numerical predictions for the impact velocities considered.

The Institute for Electrical and Electronics Engineers (IEEE), through its Standards Coordinating Committee 20 (SCC20), is developing interface standards focusing on Automatic Test System-related elements in cooperation with a Department of Defense (DoD) initiative to define, demonstrate, and recommend such standards.12 One of these standards-IEEE Std 1232–2002 Artificial Intelligence Exchange and Service Tie to All Test Environments (AI-ESTATE)-has been chosen for demonstration. Previously, we presented the results of the first phase of the AI-ESTATE demonstration, focusing on semantic interoperability of diagnostic models. The results of that demonstration successfully showed the effectiveness of semantic modeling in information exchange. In addition, the engineering burden imposed by stronger semantic requirements was demonstrated to be manageable. In the second phase, the focus was on supporting reasoner interoperability by implementing semantically defined software services in a service-oriented architecture. Here, we present an overview of the semantic interoperability problem in the context of diagnostic reasoning and discuss the results of the second phase of the demonstration.