Achieve fairness without demographics for dermatological disease diagnosis.

In medical image diagnosis, fairness has become increasingly crucial. Without bias mitigation, deploying unfair AI would harm the interests of the underprivileged population and potentially tear society apart. Recent research addresses prediction biases in deep learning models concerning demographic groups (e.g., gender, age, and race) by utilizing demographic (sensitive attribute) information during training. However, many sensitive attributes naturally exist in dermatological disease images. If the trained model only targets fairness for a specific attribute, it remains unfair for other attributes. Moreover, training a model that can accommodate multiple sensitive attributes is impractical due to privacy concerns. To overcome this, we propose a method enabling fair predictions for sensitive attributes during the testing phase without using such information during training. Inspired by prior work highlighting the impact of feature entanglement on fairness, we enhance the model features by capturing the features related to the sensitive and target attributes and regularizing the feature entanglement between corresponding classes. This ensures that the model can only classify based on the features related to the target attribute without relying on features associated with sensitive attributes, thereby improving fairness and accuracy. Additionally, we use disease masks from the Segment Anything Model (SAM) to enhance the quality of the learned feature. Experimental results demonstrate that the proposed method can improve fairness in classification compared to state-of-the-art methods in two dermatological disease datasets.

Drug screening on digital microfluidics for cancer precision medicine.

Drug screening based on in-vitro primary tumor cell culture has demonstrated potential in personalized cancer diagnosis. However, the limited number of tumor cells, especially from patients with early stage cancer, has hindered the widespread application of this technique. Hence, we developed a digital microfluidic system for drug screening using primary tumor cells and established a working protocol for precision medicine. Smart control logic was developed to increase the throughput of the system and decrease its footprint to parallelly screen three drugs on a 4 × 4 cm2 chip in a device measuring 23 × 16 × 3.5 cm3. We validated this method in an MDA-MB-231 breast cancer xenograft mouse model and liver cancer specimens from patients, demonstrating tumor suppression in mice/patients treated with drugs that were screened to be effective on individual primary tumor cells. Mice treated with drugs screened on-chip as ineffective exhibited similar results to those in the control groups. The effective drug identified through on-chip screening demonstrated consistency with the absence of mutations in their related genes determined via exome sequencing of individual tumors, further validating this protocol. Therefore, this technique and system may promote advances in precision medicine for cancer treatment and, eventually, for any disease.

GNN-based Concentration Prediction with Variable Input Flow Rates for Microfluidic Mixers.

Recent years have witnessed significant advances brought by microfluidic biochips in automating biochemical protocols. Accurate preparation of fluid samples is an essential component of these protocols, where concentration prediction and generation are critical. Equipped with the advantages of convenient fabrication and control, microfluidic mixers demonstrate huge potential in sample preparation. Although finite element analysis (FEA) is the most commonly used simulation method for accurate concentration prediction of a given microfluidic mixer, it is time-consuming with poor scalability for large biochip sizes. Recently, machine learning models have been adopted in concentration prediction, with great potential in enhancing the efficiency over traditional FEA methods. However, the state-of-the-art machine learning-based method can only predict the concentration of mixers with fixed input flow rates and fixed sizes. In this paper, we propose a new concentration prediction method based on graph neural networks (GNNs), which can predict output concentrations for microfluidic mixters with variable input flow rates. Moreover, a transfer learning method is proposed to transfer the trained model to mixers of different sizes with reduced training data. Experimental results show that, for microfluidic mixers with fixed input flow rates, the proposed method obtains an average reduction of 88% in terms of prediction errors compared with the state-of-the-art method. For microfluidic mixers with variable input flow rates, the proposed method reduces the prediction error by 85% on average. Besides, the proposed transfer learning method reduces the training data by 84% for extending the pre-trained model for microfluidic mixers of different sizes with acceptable prediction error.

Automated design of a 3D passive microfluidic particle sorter.

Microfluidic chips that can sort mixtures of cells and other particles have important applications in research and healthcare. However, designing a sorter chip for a given application is a slow and difficult process, especially when we extend the design space from 2D into a 3D scenario. Compared to the 2D scenario, we need to explore more geometries to derive the appropriate design due to the extra dimension. To evaluate sorting performance, the simulation of the particle trajectory is needed. The 3D scenario brings particle trajectory simulation more challenges of runtime and collision handling with irregular obstacle shapes. In this paper, we propose a framework to design a 3D microfluidic particle sorter for a given application with an efficient 3D particle trajectory simulator. The efficient simulator enables us to simulate more samples to ensure the robustness of the sorting performance. Our experimental result shows that the sorter designed by our framework successfully separates the particles with the targeted size.

Machine Learning for Two-Phase Flow Separation in a Liquid-Liquid Interface Manipulation Separator.

Two-phase flow separation is a key step in various downstream purification processes. The use of a separator with controllable flow behavior is recommended to avoid contamination. In this study, a core-annular separator for biphasic flow separation with four different chemical polarities was developed, and two machine learning-based methods were proposed for answering two emergent questions to meet real industrial needs. (1) Could complete two-phase separation be achieved under these operating conditions? (2) Could the separation process be accelerated by determining the maximum input flow rate of the water? Process prediction for automation, machine learning-based classifiers, and multilayer perceptron were used to address these questions by predicting successful separation and the maximum input flow rates of unknown water-solvent systems with limited experimental data as training samples. The core-annular separator achieved complete two-phase water-solvent separation at a maximum total input flow rate of 4000 µL min-1. Moreover, the classification accuracy for complete separation reached 92.2%, and the multilayer perceptron network had the best performance for predicting the flow rate. This liquid-liquid interface manipulation separator and machine learning method could decrease the cost of relevant process development.

Conquering the Tyranny of Number With Digital Microfluidics.

The development of large-scale integration based on soft lithography has ushered a new revolution in microfluidics. This technology, however, relies inherently on pneumatic control of micromechanical valves that require air pressure to operate, while digital microfluidics uses a purely electrical signal on an electrode for droplet manipulation. In this article, we discuss the prospect and current challenges of digital microfluidics to solve the problem of the tyranny of numbers in arbitrary fluidic manipulation. We distill the fundamental physics governing electrowetting and their implications for specifications of the control electronics. We survey existing control electronics in digital microfluidics and detail the improvements needed to realize a low-power, programmable digital microfluidic system. Such an instrument would attract wide interest in both professional and non-professional (hobbyist) communities.

Automated calibration of 3D-printed microfluidic devices based on computer vision.

With the development of 3D printing techniques, the application of it in microfluidic/Lab-on-a-Chip (LoC) fabrication is becoming more and more attractive. However, to achieve a satisfying printing quality of the target devices, researchers usually require quite an amount of work in calibration trials even for high-end 3D printers. To increase the calibration efficiency of the average priced printers and promote the application of 3D printing technology in the microfluidic community, this work has presented a computer vision (CV)-based method for rapid and precise 3D printing calibration with examples on cylindrical hole/post diameters of 0.2-2.4 mm and rectangular hole/post widths of 0.2-1.0 mm by a stereolithography-based 3D printer. Our method is fully automated, which contains five steps and only needs a camera at hand to provide photos for convolutional neural network recognition. The experimental results showed that our CV-based method could provide calibrated dimensions with just one print of the specific calibration ruler to meet user desire. The higher resolution of the photo provides a higher precision in calibration. Subsequently, only one more print for the target device is needed after the calibration process. Overall, this work has provided a quick and precise calibration tool for researchers to apply 3D printing in the fabrication of their microfluidic/LoC devices with average price printers. Besides, with our open source calibration software and calibration ruler design file, researchers can modify the specific setting based on customized needs and conduct calibration on any type of 3D printer.

Predicting the fluid behavior of random microfluidic mixers using convolutional neural networks.

With the various applications of microfluidics, numerical simulation is highly recommended to verify its performance and reveal potential defects before fabrication. Among all the simulation parameters and simulation tools, the velocity field and concentration profile are the key parts and are generally simulated using finite element analysis (FEA). In our previous work [Wang et al., Lab Chip, 2016, 21, 4212-4219], automated design of microfluidic mixers by pre-generating a random library with the FEA was proposed. However, the duration of the simulation process is time-consuming, while the matching consistency between limited pre-generated designs and user desire is not stable. To address these issues, we inventively transformed the fluid mechanics problem into an image recognition problem and presented a convolutional neural network (CNN)-based technique to predict the fluid behavior of random microfluidic mixers. The pre-generated 10 513 candidate designs in the random library were used in the training process of the CNN, and then 30 757 brand new microfluidic mixer designs were randomly generated, whose performance was predicted by the CNN. Experimental results showed that the CNN method could complete all the predictions in just 10 seconds, which was around 51 600× faster than the previous FEA method. The CNN library was extended to contain 41 270 candidate designs, which has filled up those empty spaces in the fluid velocity versus solute concentration map of the random library, and able to provide more choices and possibilities for user desire. Besides, the quantitative analysis has confirmed the increased compatibility of the CNN library with user desire. In summary, our CNN method not only presents a much faster way of generating a more complete library with candidate mixer designs but also provides a solution for predicting fluid behavior using a machine learning technique.

Detailed profiling of carbon fixation of in silico synthetic autotrophy with reductive tricarboxylic acid cycle and Calvin-Benson-Bassham cycle in Esherichia coli using hydrogen as an energy source.

Carbon fixation is the main route of inorganic carbon in the form of CO2 into the biosphere. In nature, RuBisCO is the most abundant protein that photosynthetic organisms use to fix CO2 from the atmosphere through the Calvin-Benson-Bassham (CBB) cycle. However, the CBB cycle is limited by its low catalytic rate and low energy efficiency. In this work, we attempt to integrate the reductive tricarboxylic acid and CBB cycles in silico to further improve carbon fixation capacity. Key heterologous enzymes, mostly carboxylating enzymes, are inserted into the Esherichia coli core metabolic network to assimilate CO2 into biomass using hydrogen as energy source. Overall, such a strain shows enhanced growth yield with simultaneous running of dual carbon fixation cycles. Our key results include the following. (i) We identified two main growth states: carbon-limited and hydrogen-limited; (ii) we identified a hierarchy of carbon fixation usage when hydrogen supply is limited; and (iii) we identified the alternative sub-optimal growth mode while performing genetic perturbation. The results and modeling approach can guide bioengineering projects toward optimal production using such a strain as a microbial cell factory.

Micro-Electrode-Dot-Array Digital Microfluidic Biochips: Technology, Design Automation, and Test Techniques.

Digital microfluidic biochips (DMFBs) are being increasingly used for DNA sequencing, point-of-care clinical diagnostics, and immunoassays. DMFBs based on a micro-electrode-dot-array (MEDA) architecture have recently been proposed, and fundamental droplet manipulations, e.g., droplet mixing and splitting, have also been experimentally demonstrated on MEDA biochips. There can be thousands of microelectrodes on a single MEDA biochip, and the fine-grained control of nanoliter volumes of biochemical samples and reagents is also enabled by this technology. MEDA biochips offer the benefits of real-time sensitivity, lower cost, easy system integration with CMOS modules, and full automation. This review paper first describes recent design tools for high-level synthesis and optimization of map bioassay protocols on a MEDA biochip. It then presents recent advances in scheduling of fluidic operations, placement of fluidic modules, droplet-size-aware routing, adaptive error recovery, sample preparation, and various testing techniques. With the help of these tools, biochip users can concentrate on the development of nanoscale bioassays, leaving details of chip optimization and implementation to software tools.

Droplet Size-Aware and Error-Correcting Sample Preparation Using Micro-Electrode-Dot-Array Digital Microfluidic Biochips.

Sample preparation in digital microfluidics refers to the generation of droplets with target concentrations for on-chip biochemical applications. In recent years, digital microfluidic biochips (DMFBs) have been adopted as a platform for sample preparation. However, there remain two major problems associated with sample preparation on a conventional DMFB. First, only a (1:1) mixing/splitting model can be used, leading to an increase in the number of fluidic operations required for sample preparation. Second, only a limited number of sensors can be integrated on a conventional DMFB; as a result, the latency for error detection during sample preparation is significant. To overcome these drawbacks, we adopt a next generation DMFB platform, referred to as micro-electrode-dot-array (MEDA), for sample preparation. We propose the first sample-preparation method that exploits the MEDA-specific advantages of fine-grained control of droplet sizes and real-time droplet sensing. Experimental demonstration using a fabricated MEDA biochip and simulation results highlight the effectiveness of the proposed sample-preparation method.

Droplet Size-Aware High-Level Synthesis for Micro-Electrode-Dot-Array Digital Microfluidic Biochips.

A digital microfluidic biochip (DMFB) is an attractive technology platform for automating laboratory procedures in biochemistry. In recent years, DMFBs based on a microelectrode-dot-array (MEDA) architecture have been demonstrated. However, due to the inherent differences between today's DMFBs and MEDA, existing synthesis solutions for biochemistry mapping cannot be utilized for MEDA biochips. We present the first synthesis approach that can be used for MEDA biochips. We first present a general analytical model for droplet velocity and validate it experimentally using a fabricated MEDA biochip. We then present the proposed synthesis method targeting reservoir placement, operation scheduling, module placement, routing of droplets of various sizes, and diagonal movement of droplets in a two-dimensional array. Simulation results using benchmarks and experimental results using a fabricated MEDA biochip demonstrate the effectiveness of the proposed synthesis technique.

Pressure-Aware Control Layer Optimization for Flow-Based Microfluidic Biochips.

Flow-based microfluidic biochips are attracting increasing attention with successful biomedical applications. One critical issue with flow-based microfluidic biochips is the large number of microvalves that require peripheral control pins. Even using the broadcasting addressing scheme, i.e., one control pin controls multiple microvalves simultaneously, thousands of microvalves would still require hundreds of control prins, which is unrealistic. To address this critical challenge in control scalability, the control-layer multiplexer is introduced to effectively reduce the number of control pins into log scale of the number of microvalves. There are two practical design issues with the control-layer multiplexer: (1) the reliability issue caused by the frequent control-valve switching, and (2) the pressure degradation problem caused by the control-valve switching without pressure refreshing from the pressure source. This paper addresses these two design issues by the proposed Hamming-distance-based switching sequence optimization method and the XOR-based pressure refreshing method. Simulation results demonstrate the effectiveness and efficiency of the proposed methods with an average 77.2% (maximum 89.6%) improvement in total pressure refreshing cost, and an average 88.5% (maximum 90.0%) improvement in pressure deviation.