Raman spectroscopy with an improved support vector machine for discrimination of thyroid and parathyroid tissues.

The objective of this study was to discriminate thyroid and parathyroid tissues using Raman spectroscopy combined with an improved support vector machine (SVM) algorithm. In thyroid surgery, there is a risk of inadvertently removing the parathyroid glands. At present, there is a lack of research on using Raman spectroscopy to discriminate parathyroid and thyroid tissues. In this article, samples were obtained from 43 individuals with thyroid and parathyroid tissues for Raman spectroscopy analysis. This study employed partial least squares (PLS) to reduce dimensions of data, and three optimization algorithms are used to improve the classification accuracy of SVM algorithm model in spectral analysis. The results show that PLS-GA-SVM algorithm has higher diagnostic accuracy and better reliability. The sensitivity of this algorithm is 94.67% and the accuracy is 94.44%. It can be concluded that Raman spectroscopy combined with the PLS-GA-SVM diagnostic algorithm has significant potential for discriminating thyroid and parathyroid tissues.

Microfluidic strategies for engineering oxygen-releasing biomaterials.

In the field of tissue engineering, local hypoxia in large-cell structures (larger than 1 mm3) poses a significant challenge. Oxygen-releasing biomaterials supply an innovative solution through oxygen ⁠ delivery in a sustained and controlled manner. Compared to traditional methods such as emulsion, sonication, and agitation, microfluidic technology offers distinct benefits for oxygen-releasing material production, including controllability, flexibility, and applicability. It holds enormous potential in the production of smart oxygen-releasing materials. This review comprehensively covers the fabrication and application of microfluidic-enabled oxygen-releasing biomaterials. To begin with, the physical mechanism of various microfluidic technologies and their differences in oxygen carrier preparation are explained. Then, the distinctions among diverse oxygen-releasing components in regards for oxygen-releasing mechanism, oxygen-carrying capacity, and duration of oxygen release are presented. Finally, the present obstacles and anticipated development trends are examined together with the application outcomes of oxygen-releasing biomaterials based on microfluidic technology in the biomedical area. STATEMENT OF SIGNIFICANCE: Oxygen is essential for sustaining life, and hypoxia (a condition of low oxygen) is a significant challenge in various diseases. Microfluidic-based oxygen-releasing biomaterials offer precise control and outstanding performance, providing unique advantages over traditional approaches for tissue engineering. However, comprehensive reviews on this topic are currently lacking. In this review, we provide a comprehensive analysis of various microfluidic technologies and their applications for developing oxygen-releasing biomaterials. We compare the characteristics of organic and inorganic oxygen-releasing biomaterials and highlight the latest advancements in microfluidic-enabled oxygen-releasing biomaterials for tissue engineering, wound healing, and drug delivery. This review may hold the potential to make a significant contribution to the field, with a profound impact on the scientific community.

Semantic uncertainty Guided Cross-Transformer for enhanced macular edema segmentation in OCT images.

Macular edema, a prevalent ocular complication observed in various retinal diseases, can lead to significant vision loss or blindness, necessitating accurate and timely diagnosis. Despite the potential of deep learning for segmentation of macular edema, challenges persist in accurately identifying lesion boundaries, especially in low-contrast and noisy regions, and in distinguishing between Inner Retinal Fluid (IRF), Sub-Retinal Fluid (SRF), and Pigment Epithelial Detachment (PED) lesions. To address these challenges, we present a novel approach, termed Semantic Uncertainty Guided Cross-Transformer Network (SuGCTNet), for the simultaneous segmentation of multi-class macular edema. Our proposed method comprises two key components, the semantic uncertainty guided attention module (SuGAM) and the Cross-Transformer module (CTM). The SuGAM module utilizes semantic uncertainty to allocate additional attention to regions with semantic ambiguity, improves the segmentation performance of these challenging areas. On the other hand, the CTM module capitalizes on both uncertainty information and multi-scale image features to enhance the overall continuity of the segmentation process, effectively minimizing feature confusion among different lesion types. Rigorous evaluation on public datasets and various OCT imaging device data demonstrates the superior performance of our proposed method compared to state-of-the-art approaches, highlighting its potential as a valuable tool for improving the accuracy and reproducibility of macular edema segmentation in clinical settings, and ultimately aiding in the early detection and diagnosis of macular edema-related diseases and associated retinal conditions.

SCAC: A Semi-Supervised Learning Approach for Cervical Abnormal Cell Detection.

Cervical abnormal cell detection plays a crucial role in the early screening of cervical cancer. In recent years, some deep learning-based methods have been proposed. However, these methods rely heavily on large amounts of annotated images, which are time-consuming and labor-intensive to acquire, thus limiting the detection performance. In this paper, we present a novel Semi-supervised Cervical Abnormal Cell detector (SCAC), which effectively utilizes the abundant unlabeled data. We utilize Transformer as the backbone of SCAC to capture long-range dependencies to mimic the diagnostic process of pathologists. In addition, in SCAC, we design a Unified Strong and Weak Augment strategy (USWA) that unifies two data augmentation pipelines, implementing consistent regularization in semi-supervised learning and enhancing the diversity of the training data. We also develop a Global Attention Feature Pyramid Network (GAFPN), which utilizes the attention mechanism to better extract multi-scale features from cervical cytology images. Notably, we have created an unlabeled cervical cytology image dataset, which can be leveraged by semi-supervised learning to enhance detection accuracy. To the best of our knowledge, this is the first publicly available large unlabeled cervical cytology image dataset. By combining this dataset with two publicly available annotated datasets, we demonstrate that SCAC outperforms other existing methods, achieving state-of-the-art performance. Additionally, comprehensive ablation studies are conducted to validate the effectiveness of USWA and GAFPN. These promising results highlight the capability of SCAC to achieve high diagnostic accuracy and extensive clinical applications.

Programmable Photoswitchable Microcapsules Enable Precise and Tailored Drug Delivery from Microfluidics.

The development of precision personalized medicine poses a significant need for the next generation of advanced diagnostic and therapeutic technologies, and one of the key challenges is the development of highly time-, space-, and dose-controllable drug delivery systems that respond to the complex physiopathology of patient populations. In response to this challenge, an increasing number of stimuli-responsive smart materials are integrated into biomaterial systems for precise targeted drug delivery. Among them, responsive microcapsules prepared by droplet microfluidics have received much attention. In this study, we present a UV-visible light cycling mediated photoswitchable microcapsule (PMC) with dynamic permeability-switching capability for precise and tailored drug release. The PMCs were fabricated using a programmable pulsed aerodynamic printing (PPAP) technique, encapsulating an aqueous core containing magnetic nanoparticles and the drug doxorubicin (DOX) within a poly(lactic-co-glycolic acid) (PLGA) composite shell modified by PEG-b-PSPA. Selective irradiation of PMCs with ultraviolet (UV) or visible light (Vis) allows for high-precision time-, space-, and dose-controlled release of the therapeutic agent. An experimentally validated theoretical model was developed to describe the drug release pattern, holding promise for future customized programmable drug release applications. The therapeutic efficacy and value of patternable cancer cell treatment activated by UV radiation is demonstrated by our experimental results. After in vitro transcatheter arterial chemoembolization (TACE), PMCs can be removed by external magnetic fields to mitigate potential side effects. Our findings demonstrate that PMCs have the potential to integrate embolization, on-demand drug delivery, magnetic actuation, and imaging properties, highlighting their immense potential for tailored drug delivery and embolic therapy.

Edge-Guided Contrastive Adaptation Network for Arteriovenous Nicking Classification Using Synthetic Data.

Retinal arteriovenous nicking (AVN) manifests as a reduced venular caliber of an arteriovenous crossing. AVNs are signs of many systemic, particularly cardiovascular diseases. Studies have shown that people with AVN are twice as likely to have a stroke. However, AVN classification faces two challenges. One is the lack of data, especially AVNs compared to the normal arteriovenous (AV) crossings. The other is the significant intra-class variations and minute inter-class differences. AVNs may look different in shape, scale, pose, and color. On the other hand, the AVN could be different from the normal AV crossing only by slight thinning of the vein. To address these challenges, first, we develop a data synthesis method to generate AV crossings, including normal and AVNs. Second, to mitigate the domain shift between the synthetic and real data, an edge-guided unsupervised domain adaptation network is designed to guide the transfer of domain invariant information. Third, a semantic contrastive learning branch (SCLB) is introduced and a set of semantically related images, as a semantic triplet, are input to the network simultaneously to guide the network to focus on the subtle differences in venular width and to ignore the differences in appearance. These strategies effectively mitigate the lack of data, domain shift between synthetic and real data, and significant intra- but minute inter-class differences. Extensive experiments have been performed to demonstrate the outstanding performance of the proposed method.

Free-Boundary Microfluidic Platform for Advanced Materials Manufacturing and Applications.

Microfluidics, with its remarkable capacity to manipulate fluids and droplets at the microscale, has emerged as a powerful platform in numerous fields. In contrast to conventional closed microchannel microfluidic systems, free-boundary microfluidic manufacturing (FBMM) processes continuous precursor fluids into jets or droplets in a relatively spacious environment. FBMM is highly regarded for its superior flexibility, stability, economy, usability, and versatility in the manufacturing of advanced materials and architectures. In this review, a comprehensive overview of recent advancements in FBMM is provided, encompassing technical principles, advanced material manufacturing, and their applications. FBMM is categorized based on the foundational mechanisms, primarily comprising hydrodynamics, interface effects, acoustics, and electrohydrodynamic. The processes and mechanisms of fluid manipulation are thoroughly discussed. Additionally, the manufacturing of advanced materials in various dimensions ranging from zero-dimensional to three-dimensional, as well as their diverse applications in material science, biomedical engineering, and engineering are presented. Finally, current progress is summarized and future challenges are prospected. Overall, this review highlights the significant potential of FBMM as a powerful tool for advanced materials manufacturing and its wide-ranging applications.

Erratum to "A Low-Cost High-Performance Data Augmentation for Deep Learning-Based Skin Lesion Classification".

[This corrects the article DOI: 10.34133/2022/9765307.].

High precision vibration sectioning for 3D imaging of the whole central nervous system.

BACKGROUND: Imaging and reconstruction of the morphology of neurons within the entire central nervous system (CNS) is important for deciphering the neural circuitry and related brain functions. With combination of tissue clearing and light sheet microscopy, previous studies have imaged the mouse CNS at cellular resolution, while remaining single axons unresolvable due to the tradeoff between sample size and imaging resolution. This could be improved by sectioning the sample into thick slices and imaged with high resolution light sheet microscopy as described in our previous study. However, the achievable quality for 3D imaging of serial thick slices is often hindered by surface undulation and other artifacts introduced by sectioning and handling limitations. NEW METHODS: In order to improve the imaging quality for mouse CNS, we develop a high-performance vibratome system for sample sectioning and handling automation. The sectioning mechanism of the system was modeled theoretically and verified experimentally. The effects of process parameters and sample properties on sectioning accuracy were studied to optimize the sectioning outcome. The resultant imaging outcome was demonstrated on mouse samples. RESULTS: Our theoretical model of vibratome effectively depicts the relationship between the sample surface undulation errors and the sectioning parameters. With the guidance of the theoretical model, the vibratome is able to achieve a local surface undulation error of ±0.5 µm and a surface arithmetic mean deviation (Sa) of 220 nm for 300-µm-thick tissue slices. Imaging results of mouse CNS show the continuous sectioning capability of the vibratome. COMPARISON WITH EXISTING METHOD: Our automatic sectioning and handling system is able to process serial thick slices for 3D imaging of the whole CNS at a single-axon resolution, superior to the commercially available vibratome devices. CONCLUSION: Our automatic sectioning and handling system can be optimized to prepare thick sample slices with minimal surface undulation and manual manipulation in support of 3D brain mapping with high-throughput and high-accuracy.

Modality preserving U-Net for segmentation of multimodal medical images.

Background: Recent advances in artificial intelligence and digital image processing have inspired the use of deep neural networks for segmentation tasks in multimodal medical imaging. Unlike natural images, multimodal medical images contain much richer information regarding different modal properties and therefore present more challenges for semantic segmentation. However, there is no report on systematic research that integrates multi-scaled and structured analysis of single-modal and multimodal medical images. Methods: We propose a deep neural network, named as Modality Preserving U-Net (MPU-Net), for modality-preserving analysis and segmentation of medical targets from multimodal medical images. The proposed MPU-Net consists of a modality preservation encoder (MPE) module that preserves the feature independency among the modalities and a modality fusion decoder (MFD) module that performs a multiscale feature fusion analysis for each modality in order to provide a rich feature representation for the final task. The effectiveness of such a single-modal preservation and multimodal fusion feature extraction approach is verified by multimodal segmentation experiments and an ablation study using brain tumor and prostate datasets from Medical Segmentation Decathlon (MSD). Results: The segmentation experiments demonstrated the superiority of MPU-Net over other methods in the segmentation tasks for multimodal medical images. In the brain tumor segmentation tasks, the Dice scores (DSCs) for the whole tumor (WT), the tumor core (TC) and the enhancing tumor (ET) regions were 89.42%, 86.92%, and 84.59%, respectively. In the meanwhile, the 95% Hausdorff distance (HD95) results were 3.530, 4.899 and 2.555, respectively. In the prostate segmentation tasks, the DSCs for the peripheral zone (PZ) and the transitional zone (TZ) of the prostate were 71.20% and 90.38%, respectively. In the meanwhile, the 95% HD95 results were 6.367 and 4.766, respectively. The ablation study showed that the combination of single-modal preservation and multimodal fusion methods improved the performance of multimodal medical image feature analysis. Conclusions: In the segmentation tasks using brain tumor and prostate datasets, the MPU-Net method has achieved the improved performance in comparison with the conventional methods, indicating its potential application for other segmentation tasks in multimodal medical images.

Oxygen-releasing biomaterials for regenerative medicine.

Oxygen is critical to the survival, function and fate of mammalian cells. Oxygen tension controls cellular behavior through metabolic programming, which in turn controls tissue regeneration. A variety of biomaterials with oxygen-releasing capabilities have been developed to provide oxygen supply to ensure cell survival and differentiation for therapeutic efficacy, and to prevent hypoxia-induced tissue damage and cell death. However, controlling the oxygen release with spatial and temporal accuracy is still technically challenging. In this review, we provide a comprehensive overview of organic and inorganic materials available as oxygen sources, including hemoglobin-based oxygen carriers (HBOCs), perfluorocarbons (PFCs), photosynthetic organisms, solid and liquid peroxides, and some of the latest materials such as metal-organic frameworks (MOFs). Additionally, we introduce the corresponding carrier materials and the oxygen production methods and present state-of-the-art applications and breakthroughs of oxygen-releasing materials. Furthermore, we discuss the current challenges and the future perspectives in the field. After reviewing the recent progress and the future perspectives of oxygen-releasing materials, we predict that smart material systems that combine precise detection of oxygenation and adaptive control of oxygen delivery will be the future trend for oxygen-releasing materials in regenerative medicine.

Converting a probe-based fluorescence system into an easy-to-use adjunct for the detection of parathyroid glands accidentally resected intraoperatively.

PURPOSE: The reported threshold of a near-infrared fluorescence detection probe (FDP) for judging parathyroid glands (PGs) is based on the autofluorescence intensity relative to other non-PG tissues, making it unreliable when not enough reference tissues are measured. We aim to convert FDP into a more convenient tool for identifying accidentally resected PGs by quantitative measurements of autofluorescence in resected tissues. METHODS: It was a prospective study approved by the Institutional Review Board. The research was divided into two stages: (1) In order to calibrate the novel FDP system, autofluorescence intensity of different in / ex vivo tissues was measured and the optimal threshold was obtained using receiver operating characteristic (ROC) curve. (2) To further validate the effectiveness of the new system, detection rates of incidental resected PGs by pathology in the control group and by FDP in the experimental group were compared. RESULTS: Autofluorescence of PGs was significantly higher than that of non-PG tissue (43 patients, Mann-Whitney U test, p < 0.0001). An optimal threshold of sensitivity / specificity (78.8% and 85.1%) for discriminating PGs was obtained. The detection rates of experimental group (20 patients) and control group (33 patients) are 5.0% and 6.1% respectively (one-tailed Fisher's exact test, p = 0.6837), indicating the novel FDP system can achieve a similar proportion of PG detection compared with pathological examinations. CONCLUSIONS: The novel FDP system can be used as an easy-to-use adjunct for detecting PG accidentally resected intraoperatively before the tissues are sent for frozen sections during thyroidectomy surgeries. TRIAL REGISTRATION: Registration number: ChiCTR2200057957.

Stereotactic co-axial projection imaging for augmented reality neuronavigation: a proof-of-concept study.

Background: Lack of intuitiveness and poor hand-eye coordination present a major technical challenge in neurosurgical navigation. Methods: We developed an integrated dexterous stereotactic co-axial projection imaging (sCPI) system featuring orthotopic image projection for augmented reality (AR) neurosurgical navigation. The performance characteristics of the sCPI system, including projection resolution and navigation accuracy, were quantitatively verified. The resolution of the sCPI was tested with a USAF1951 resolution test chart. The stereotactic navigation accuracy of the sCPI was measured using a calibration panel with a 7×7 circle array pattern. In benchtop validation, the navigation accuracy of the sCPI and the BrainLab Kick Navigation Station was compared using a skull phantom with 8 intracranial targets. Finally, we demonstrated the potential clinical application of sCPI through a clinical trial. Results: The resolution test showed that the resolution of the sCPI was 1.3 mm. In a stereotactic navigation accuracy test, the maximum and minimum error of the sCPI was 2.9 and 0.3 mm, and the mean error was 1.5 mm. The stereotactic navigation accuracy test also showed that the navigation error of the sCPI would increase with the pitch and yaw angle, but there was no obvious difference in navigation errors caused by different yaw directions, which meant that the navigation error is unbiased across all directions. The benchtop validation showed that the average navigation errors for the sCPI system and the Kick Navigation Station were 1.4±0.8 and 1.8±0.7 mm, the medians were 1.3 and 1.9 mm, and the average preparation times were 3 min 24 sec and 6 min 8 sec, respectively. The clinical feasibility of sCPI-assisted neurosurgical navigation was demonstrated in a clinical study. In comparison with the BrainLab device, the sCPI system required less time for preoperative preparation and enhanced the clinician experience in intraoperative visualization and navigation. Conclusions: The sCPI technique can be potentially used in many surgical applications for intuitive visualization of medical information and intraoperative guidance of surgical trajectories.

Co-axial Projective Imaging for Augmented Reality Telementoring in Skin Cancer Surgery.

Telemedicine has the potential to overcome the unequal distribution of medical resources worldwide. In this study, we report the second-generation co-axial projective imaging (CPI-2) system featured with orthotopic image projection for augmented reality surgical telementoring. The CPI-2 system can acquire surgical scene images from the local site, transmit them wirelessly to the remote site, and project the virtual annotations drawn by a remote expert with great accuracy to the surgical field. The performance characteristics of the CPI-2 system are quantitatively verified in benchtop experiments. The ex vivo study that compares the CPI-2 system and a monitor-based telementoring system shows that the CPI-2 system can reduce the focus shift and avoid subjective mapping of the instructions from a monitor to the real-world scene, thereby saving operation time and achieving precise teleguidance. The clinical feasibility of the CPI-2 system is validated in teleguided skin cancer surgery. Our ex vivo and in vivo experiment results imply the improved performance of surgical telementoring, and the clinical utility of deploying the CPI-2 system for surgical interventions in resource-limited settings. The CPI-2 system has the potential to reduce healthcare disparities in remote areas with limited resources.

Autofluorescence detection and co-axial projection for intraoperative localization of parathyroid gland.

BACKGROUND: Near-infrared (NIR) autofluorescence detection is an effective method for identifying parathyroid glands (PGs) in thyroidectomy or parathyroidectomy. Fiber optical probes provide quantitative autofluorescence measurements for PG detection owing to its high sensitivity and high excitation light cut-off efficiency at a fixed detection distance. However, an optical fiber probe lacks the imaging capability and cannot map the autofluorescence distribution on top of normal tissue background. Therefore, there is a need for intraoperative mapping of PGs with high sensitivity and imaging resolution. METHODS: We have developed a fluorescence scanning and projection (FSP) system that combines a scanning probe and a co-axial projector for intraoperative localization and in situ display of PGs. Some of the key performance characteristics, including spatial resolution and sensitivity for detection, spatial resolution for imaging, dynamic time latency, and PG localization capability, are characterized and verified by benchtop experiments. Clinical utility of the system is simulated by a fluorescence-guided PG localization surgery on a tissue-simulating phantom and validated in an ex vivo experiment. RESULTS: The system is able to detect indocyanine green (ICG) solution of 5 pM at a high signal-to-noise ratio (SNR). Additionally, it has a maximal projection error of 0.92 mm, an averaged projection error of 0.5 ± 0.23 mm, and an imaging resolution of 748 µm at a working distance ranging from 35 to 55 cm. The dynamic testing yields a short latency of 153 ± 54 ms, allowing for intraoperative scanning on target tissue during a surgical intervention. The simulated fluorescence-guided PG localization surgery has validated the system's capability to locate PG phantom with operating room ambient light interference. The simulation experiment on the PG phantom yields a position detection bias of 0.36 ± 0.17 mm, and an area intersection over unit (IoU) of 76.6% ± 6.4%. Fluorescence intensity attenuates exponentially with the thickness of covered tissue over the PG phantom, indicating the need to remove surrounding tissue in order to reveal the weak autofluorescence signal from PGs. The ex vivo experiment demonstrates the technical feasibility of the FSP system for intraoperative PG localization with accuracy. CONCLUSION: We have developed a novel probe-based imaging and navigation system with high sensitivity for fluorescence detection, capability for fluorescence image reconstruction, multimodal image fusion and in situ PG display function. Our studies have demonstrated its clinical potential for intraoperative localization and in situ display of PGs in thyroidectomy or parathyroidectomy.

Low-cost thermal imaging with machine learning for non-invasive diagnosis and therapeutic monitoring of pneumonia.

Rapid screening and early treatment of lung infection are essential for effective control of many epidemics such as Coronavirus Disease 2019 (COVID-19). Recent studies have demonstrated the potential correlation between lung infection and the change of back skin temperature distribution. Based on these findings, we propose to use low-cost, portable and rapid thermal imaging in combination with image-processing algorithms and machine learning analysis for non-invasive and safe detection of pneumonia. The proposed method was tested in 69 subjects (30 normal adults, 11 cases of fever without pneumonia, 19 cases of general pneumonia and 9 cases of COVID-19) where both RGB and thermal images were acquired from the back of each subject. The acquired images were processed automatically in order to extract multiple location and shape features that distinguish normal subjects from pneumonia patients at a high accuracy of 93 % . Furthermore, daily assessment of two pneumonia patients by the proposed method accurately predicted the clinical outcomes, coincident with those of laboratory tests. Our pilot study demonstrated the technical feasibility of portable and intelligent thermal imaging for screening and therapeutic assessment of pneumonia. The method can be potentially implemented in under-resourced regions for more effective control of respiratory epidemics.

A Single Model Deep Learning Approach for Alzheimer's Disease Diagnosis.

Early and accurate diagnosis of Alzheimer's disease (AD) and its prodromal period mild cognitive impairment (MCI) is essential for the delayed disease progression and the improved quality of patients' life. The emerging computer-aided diagnostic methods that combine deep learning with structural magnetic resonance imaging (sMRI) have achieved encouraging results, but some of them are limit of issues such as data leakage, overfitting, and unexplainable diagnosis. In this research, we propose a novel end-to-end deep learning approach for automated diagnosis of AD. This approach has the following differences from the current approaches: (1) Convolutional Neural Network (CNN) models of different structures and capacities are evaluated systemically and the most suitable model is adopted for AD diagnosis; (2) A data augmentation strategy named Two-stage Random RandAugment (TRRA) is proposed to alleviate the overfitting issue caused by limited training data and to improve the classification performance in AD diagnosis; (3) An explainable method of Grad-CAM++ is introduced to generate the visually explainable heatmaps to make our model more transparent. Our approach has been evaluated on two publicly accessible datasets for two classification tasks of AD vs. cognitively normal (CN) and progressive MCI (pMCI) vs. stable MCI (sMCI). The experimental results indicate that our approach outperforms the state-of-the-art approaches, including those using multi-model and three-dimensional (3D) CNN methods. The resultant heatmaps from our approach also highlight the lateral ventricle and some regions of cortex, which have been proved to be affected by AD.

A Low-Cost High-Performance Data Augmentation for Deep Learning-Based Skin Lesion Classification.

Objective and Impact Statement. There is a need to develop high-performance and low-cost data augmentation strategies for intelligent skin cancer screening devices that can be deployed in rural or underdeveloped communities. The proposed strategy can not only improve the classification performance of skin lesions but also highlight the potential regions of interest for clinicians' attention. This strategy can also be implemented in a broad range of clinical disciplines for early screening and automatic diagnosis of many other diseases in low resource settings. Methods. We propose a high-performance data augmentation strategy of search space 101, which can be combined with any model through a plug-and-play mode and search for the best argumentation method for a medical database with low resource cost. Results. With EfficientNets as a baseline, the best BACC of HAM10000 is 0.853, outperforming the other published models of "single-model and no-external-database" for ISIC 2018 Lesion Diagnosis Challenge (Task 3). The best average AUC performance on ISIC 2017 achieves 0.909 (±0.015), exceeding most of the ensembling models and those using external datasets. Performance on Derm7pt archives the best BACC of 0.735 (±0.018) ahead of all other related studies. Moreover, the model-based heatmaps generated by Grad-CAM++ verify the accurate selection of lesion features in model judgment, further proving the scientific rationality of model-based diagnosis. Conclusion. The proposed data augmentation strategy greatly reduces the computational cost for clinically intelligent diagnosis of skin lesions. It may also facilitate further research in low-cost, portable, and AI-based mobile devices for skin cancer screening and therapeutic guidance.

Single Model Deep Learning on Imbalanced Small Datasets for Skin Lesion Classification.

Deep convolutional neural network (DCNN) models have been widely explored for skin disease diagnosis and some of them have achieved the diagnostic outcomes comparable or even superior to those of dermatologists. However, broad implementation of DCNN in skin disease detection is hindered by small size and data imbalance of the publically accessible skin lesion datasets. This paper proposes a novel single-model based strategy for classification of skin lesions on small and imbalanced datasets. First, various DCNNs are trained on different small and imbalanced datasets to verify that the models with moderate complexity outperform the larger models. Second, regularization DropOut and DropBlock are added to reduce overfitting and a Modified RandAugment augmentation strategy is proposed to deal with the defects of sample underrepresentation in the small dataset. Finally, a novel Multi-Weighted New Loss (MWNL) function and an end-to-end cumulative learning strategy (CLS) are introduced to overcome the challenge of uneven sample size and classification difficulty and to reduce the impact of abnormal samples on training. By combining Modified RandAugment, MWNL and CLS, our single DCNN model method achieved the classification accuracy comparable or superior to those of multiple ensembling models on different dermoscopic image datasets. Our study shows that this method is able to achieve a high classification performance at a low cost of computational resources and inference time, potentially suitable to implement in mobile devices for automated screening of skin lesions and many other malignancies in low resource settings.