Deep Learning and Multimodal Artificial Intelligence in Orthopaedic Surgery.

This review article focuses on the applications of deep learning with neural networks and multimodal neural networks in the orthopaedic domain. By providing practical examples of how artificial intelligence (AI) is being applied successfully in orthopaedic surgery, particularly in the realm of imaging data sets and the integration of clinical data, this study aims to provide orthopaedic surgeons with the necessary tools to not only evaluate existing literature but also to consider AI's potential in their own clinical or research pursuits. We first review standard deep neural networks which can analyze numerical clinical variables, then describe convolutional neural networks which can analyze image data, and then introduce multimodal AI models which analyze various types of different data. Then, we contrast these deep learning techniques with related but more limited techniques such as radiomics, describe how to interpret deep learning studies, and how to initiate such studies at your institution. Ultimately, by empowering orthopaedic surgeons with the knowledge and know-how of deep learning, this review aspires to facilitate the translation of research into clinical practice, thereby enhancing the efficacy and precision of real-world orthopaedic care for patients.

Tumor Board Shadowing: A Unique Approach for Integrating Radiation Oncologists Into General Medical Student Education.

PURPOSE: Radiation oncology is often overlooked in US medical school curricula, with few opportunities for most students to learn about the specialty or the value of radiation therapy in cancer care. Tumor boards represent a potential avenue not only to increase students' exposure to radiation oncologists but also to provide a fundamental understanding of the multidisciplinary nature of cancer care and effective collaboration in clinical practice. METHODS AND MATERIALS: In this study, we evaluated a novel radiation oncologist-driven tumor board shadowing experience at 3 medical schools in the United States and Canada. A total of 323 first- and second-year medical students participated, of whom 77.4% completed a follow-up survey assessing the effectiveness of the program as a learning tool. RESULTS: Compared with traditional clinical shadowing, students were more likely to believe that tumor board shadowing provided a similar or better experience in terms of educational content (85%), exposure to a new field (96%), and overall experience (89%). Forty-eight percent of students perceived a greater amount of multidisciplinary collaboration in oncologic care than they thought existed prior to attending. Forty-eight percent of students also felt more competent interacting with oncologists after participating, whereas 21% felt more competent interacting with patients with cancer. Students' perception of increased competence was correlated with the amount of time their assigned physician mentor spent answering their questions after the tumor board (P < .01). Second-year medical students also had a more favorable overall experience than first-year medical students did (P = .04). CONCLUSIONS: Multidisciplinary tumor boards can be used effectively as a unique immersive learning opportunity that can be feasibly implemented to improve knowledge of clinical oncology and multidisciplinary care in medical schools and expose students to physicians in smaller fields such as radiation oncology.

Contrast sensitivity of MT receptive field centers and surrounds.

Neurons throughout the visual system have receptive fields with both excitatory and suppressive components. The latter are responsible for a phenomenon known as surround suppression, in which responses decrease as a stimulus is extended beyond a certain size. Previous work has shown that surround suppression in the primary visual cortex depends strongly on stimulus contrast. Such complex center-surround interactions are thought to relate to a variety of functions, although little is known about how they affect responses in the extrastriate visual cortex. We have therefore examined the interaction of center and surround in the middle temporal (MT) area of the macaque (Macaca mulatta) extrastriate cortex by recording neuronal responses to stimuli of different sizes and contrasts. Our findings indicate that surround suppression in MT is highly contrast dependent, with the strongest suppression emerging unexpectedly at intermediate stimulus contrasts. These results can be explained by a simple model that takes into account the nonlinear contrast sensitivity of the neurons that provide input to MT. The model also provides a qualitative link to previous reports of a topographic organization of area MT based on clusters of neurons with differing surround suppression strength. We show that this organization can be detected in the gamma-band local field potentials (LFPs) and that the model parameters can predict the contrast sensitivity of these LFP responses. Overall our results show that surround suppression in area MT is far more common than previously suspected, highlighting the potential functional importance of the accumulation of nonlinearities along the dorsal visual pathway.

Perceptual and neural consequences of rapid motion adaptation.

Nervous systems adapt to the prevailing sensory environment, and the consequences of this adaptation can be observed in the responses of single neurons and in perception. Given the variety of timescales underlying events in the natural world, determining the temporal characteristics of adaptation is important to understanding how perception adjusts to its sensory environment. Previous work has shown that neural adaptation can occur on a timescale of milliseconds, but perceptual adaptation has generally been studied over relatively long timescales, typically on the order of seconds. This disparity raises important questions. Can perceptual adaptation be observed at brief, functionally relevant timescales? And if so, how do its properties relate to the rapid adaptation seen in cortical neurons? We address these questions in the context of visual motion processing, a perceptual modality characterized by rapid temporal dynamics. We demonstrate objectively that 25 ms of motion adaptation is sufficient to generate a motion aftereffect, an illusory sensation of movement experienced when a moving stimulus is replaced by a stationary pattern. This rapid adaptation occurs regardless of whether the adapting motion is perceived. In neurophysiological recordings from the middle temporal area of primate visual cortex, we find that brief motion adaptation evokes direction-selective responses to subsequently presented stationary stimuli. A simple model shows that these neural responses can explain the consequences of rapid perceptual adaptation. Overall, we show that the motion aftereffect is not merely an intriguing perceptual illusion, but rather a reflection of rapid neural and perceptual processes that can occur essentially every time we experience motion.

The role of V1 surround suppression in MT motion integration.

Neurons in the primate extrastriate cortex are highly selective for complex stimulus features such as faces, objects, and motion patterns. One explanation for this selectivity is that neurons in these areas carry out sophisticated computations on the outputs of lower-level areas such as primary visual cortex (V1), where neuronal selectivity is often modeled in terms of linear spatiotemporal filters. However, it has long been known that such simple V1 models are incomplete because they fail to capture important nonlinearities that can substantially alter neuronal selectivity for specific stimulus features. Thus a key step in understanding the function of higher cortical areas is the development of realistic models of their V1 inputs. We have addressed this issue by constructing a computational model of the V1 neurons that provide the strongest input to extrastriate cortical middle temporal (MT) area. We find that a modest elaboration to the standard model of V1 direction selectivity generates model neurons with strong end-stopping, a property that is also found in the V1 layers that provide input to MT. With this computational feature in place, the seemingly complex properties of MT neurons can be simulated by assuming that they perform a simple nonlinear summation of their inputs. The resulting model, which has a very small number of free parameters, can simulate many of the diverse properties of MT neurons. In particular, we simulate the invariance of MT tuning curves to the orientation and length of tilted bar stimuli, as well as the accompanying temporal dynamics. We also show how this property relates to the continuum from component to pattern selectivity observed when MT neurons are tested with plaids. Finally, we confirm several key predictions of the model by recording from MT neurons in the alert macaque monkey. Overall our results demonstrate that many of the seemingly complex computations carried out by high-level cortical neurons can in principle be understood by examining the properties of their inputs.

Pattern motion selectivity of spiking outputs and local field potentials in macaque visual cortex.

The dorsal pathway of the primate visual cortex is involved in the processing of motion signals that are useful for perception and behavior. Along this pathway, motion information is first measured by the primary visual cortex (V1), which sends specialized projections to extrastriate regions such as the middle temporal area (MT). Previous work with plaid stimuli has shown that most V1 neurons respond to the individual components of moving stimuli, whereas some MT neurons are capable of estimating the global motion of the pattern. In this work, we show that the majority of neurons in the medial superior temporal area (MST), which receives input from MT, have this pattern-selective property. Interestingly, the local field potentials (LFPs) measured simultaneously with the spikes often exhibit properties similar to that of the presumptive feedforward input to each area: in the high-gamma frequency band, the LFPs in MST are as component selective as the spiking outputs of MT, and MT LFPs have plaid responses that are similar to the spiking outputs of V1. In the lower LFP frequency bands (beta and low gamma), component selectivity is very common, and pattern selectivity is almost entirely absent in both MT and MST. Together, these results suggest a surprisingly strong link between the sensory tuning of cortical LFPs and afferent inputs, with important implications for the interpretation of imaging studies and for models of cortical function.