Testing the generalizability and effectiveness of deep learning models among clinics: sperm detection as a pilot study.

BACKGROUND: Deep learning has been increasingly investigated for assisting clinical in vitro fertilization (IVF). The first technical step in many tasks is to visually detect and locate sperm, oocytes, and embryos in images. For clinical deployment of such deep learning models, different clinics use different image acquisition hardware and different sample preprocessing protocols, raising the concern over whether the reported accuracy of a deep learning model by one clinic could be reproduced in another clinic. Here we aim to investigate the effect of each imaging factor on the generalizability of object detection models, using sperm analysis as a pilot example. METHODS: Ablation studies were performed using state-of-the-art models for detecting human sperm to quantitatively assess how model precision (false-positive detection) and recall (missed detection) were affected by imaging magnification, imaging mode, and sample preprocessing protocols. The results led to the hypothesis that the richness of image acquisition conditions in a training dataset deterministically affects model generalizability. The hypothesis was tested by first enriching the training dataset with a wide range of imaging conditions, then validated through internal blind tests on new samples and external multi-center clinical validations. RESULTS: Ablation experiments revealed that removing subsets of data from the training dataset significantly reduced model precision. Removing raw sample images from the training dataset caused the largest drop in model precision, whereas removing 20x images caused the largest drop in model recall. by incorporating different imaging and sample preprocessing conditions into a rich training dataset, the model achieved an intraclass correlation coefficient (ICC) of 0.97 (95% CI: 0.94-0.99) for precision, and an ICC of 0.97 (95% CI: 0.93-0.99) for recall. Multi-center clinical validation showed no significant differences in model precision or recall across different clinics and applications. CONCLUSIONS: The results validated the hypothesis that the richness of data in the training dataset is a key factor impacting model generalizability. These findings highlight the importance of diversity in a training dataset for model evaluation and suggest that future deep learning models in andrology and reproductive medicine should incorporate comprehensive feature sets for enhanced generalizability across clinics.

Modeling and Compensation of Positioning Error in Micromanipulation.

In order to improve the positioning accuracy of the micromanipulation system, a comprehensive error model is first established to take into account the microscope nonlinear imaging distortion, camera installation error, and the mechanical displacement error of the motorized stage. A novel error compensation method is then proposed with distortion compensation coefficients obtained by the Levenberg-Marquardt optimization algorithm combined with the deduced nonlinear imaging model. The compensation coefficients for camera installation error and mechanical displacement error are derived from the rigid-body translation technique and image stitching algorithm. To validate the error compensation model, single shot and cumulative error tests were designed. The experimental results show that after the error compensation, the displacement errors were controlled within 0.25 µm when moving in a single direction and within 0.02 µm per 1000 µm when moving in multiple directions.

Design and Performance Evaluation of a Wearable Sensing System for Lower-Limb Exoskeleton.

Because the target users of the assistive-type lower extremity exoskeletons (ASLEEs) are those who suffer from lower limb disabilities, customized gait is adopted for the control of ASLEEs. However, the customized gait is unable to provide stable motion for variable terrain, for example, flat, uphill, downhill, and soft ground. The purpose of this paper is to realize gait detection and environment feature recognition for AIDER by developing a novel wearable sensing system. The wearable sensing system employs 7 force sensors as a sensing matrix to achieve high accuracy of ground reaction force detection. There is one more IMU sensor that is integrated into the structure to detect the angular velocity. By fusing force and angular velocity data, four typical terrain features can be recognized successfully, and the recognition rate can reach up to 93%.

Dynamic Balance Gait for Walking Assistance Exoskeleton.

PURPOSE: Powered lower-limb exoskeleton has gained considerable interests, since it can help patients with spinal cord injury(SCI) to stand and walk again. Providing walking assistance with SCI patients, most exoskeletons are designed to follow predefined gait trajectories, which makes the patient walk unnaturally and feels uncomfortable. Furthermore, exoskeletons with predefined gait trajectories cannot always maintain balance walking especially when encountering disturbances. DESIGN/METHODOLOGY/APPROACH: This paper proposed a novel gait planning approach, which aims to provide reliable and balance gait during walking assistance. In this approach, we model the exoskeleton and patient together as a linear inverted pendulum (LIP) and obtain the patients intention through orbital energy diagram. To achieve dynamic gait planning of exoskeleton, the dynamic movement primitive (DMP) is utilized to model the gait trajectory. Meanwhile, the parameters of DMP are updated dynamically during one step, which aims to improve the ability of counteracting external disturbance. FINDINGS: The proposed approach is validated in a human-exoskeleton simulation platform, and the experimental results show the effectiveness and advantages of the proposed approach. ORIGINALITY/VALUE: We decomposed the issue of obtain dynamic balance gait into three parts: (1) based on the sensory information of exoskeleton, the intention estimator is designed to estimate the intention of taking a step; (2) at the beginning of each step, the discrete gait planner utilized the obtained gait parameters such as step length S and step duration T and generate the trajectory of swing foot based on (S, T); (3) during walking process, continuous gait regulator is utilized to adjust the gait generated by discrete gait planner to counteract disturbance.