Automated size-specific dose estimates using deep learning image processing

An automated vendor-independent system for dose monitoring in computed tomography (CT) medical examinations involving ionizing radiation is presented in this paper. The system provides precise size-specific dose estimates (SSDE) following the American Association of Physicists in Medicine regulations. Our dose management can operate on incomplete DICOM header metadata by retrieving necessary information from the dose report image by using optical character recognition. For the determination of the patient’s effective diameter and water equivalent diameter, a convolutional neural network is employed for the semantic segmentation of the body area in axial CT slices. Validation experiments for the assessment of the SSDE determination and subsequent stages of our methodology involved a total of 335 CT series (60 352 images) from both public databases and our clinical data. We obtained the mean body area segmentation accuracy of 0.9955 and Jaccard index of 0.9752, yielding a slice-wise mean absolute error of effective diameter below 2 mm and water equivalent diameter at 1 mm, both below 1%. Three modes of the SSDE determination approach were investigated and compared to the results provided by the commercial system GE DoseWatch in three different body region categories: head, chest, and abdomen. Statistical analysis was employed to point out some significant remarks, especially in the head category.     

MicroHikari3D: an automated DIY digital microscopy platform with deep learning capabilities

A microscope is an essential tool in biosciences and production quality laboratories for unveiling the secrets of microworlds. This paper describes the development of MicroHikari3D, an affordable DIY optical microscopy platform with automated sample positioning, autofocus and several illumination modalities to provide a high-quality flexible microscopy tool for labs with a short budget. This proposed optical microscope design aims to achieve high customization capabilities to allow whole 2D slide imaging and observation of 3D live specimens. The MicroHikari3D motion control system is based on the entry level 3D printer kit Tronxy X1 controlled from a server running in a Raspberry Pi 4. The server provides services to a client mobile app for video/image acquisition, processing, and a high level classification task by applying deep learning models.     

Cross-dimensional transfer learning in medical image segmentation with deep learning

Over the last decade, convolutional neural networks have emerged and advanced the state-of-the-art in various image analysis and computer vision applications. The performance of 2D image classification networks is constantly improving and being trained on databases made of millions of natural images. Conversely, in the field of medical image analysis, the progress is also remarkable but has mainly slowed down due to the relative lack of annotated data and besides, the inherent constraints related to the acquisition process. These limitations are even more pronounced given the volumetry of medical imaging data. In this paper, we introduce an efficient way to transfer the efficiency of a 2D classification network trained on natural images to 2D, 3D uni- and multi-modal medical image segmentation applications. In this direction, we designed novel architectures based on two key principles: weight transfer by embedding a 2D pre-trained encoder into a higher dimensional U-Net, and dimensional transfer by expanding a 2D segmentation network into a higher dimension one. The proposed networks were tested on benchmarks comprising different modalities: MR, CT, and ultrasound images. Our 2D network ranked first on the CAMUS challenge dedicated to echo-cardiographic data segmentation and surpassed the state-of-the-art. Regarding 2D/3D MR and CT abdominal images from the CHAOS challenge, our approach largely outperformed the other 2D-based methods described in the challenge paper on Dice, RAVD, ASSD, and MSSD scores and ranked third on the online evaluation platform. Our 3D network applied to the BraTS 2022 competition also achieved promising results, reaching an average Dice score of 91.69% (91.22%) for the whole tumor, 83.23% (84.77%) for the tumor core and 81.75% (83.88%) for enhanced tumor using the approach based on weight (dimensional) transfer. Experimental and qualitative results illustrate the effectiveness of our methods for multi-dimensional medical image segmentation.     

Multi-site cross-organ calibrated deep learning (MuSClD): Automated diagnosis of non-melanoma skin cancer

Although deep learning (DL) has demonstrated impressive diagnostic performance for a variety of computational pathology tasks, this performance often markedly deteriorates on whole slide images (WSI) generated at external test sites. This phenomenon is due in part to domain shift, wherein differences in test-site pre-analytical variables (e.g., slide scanner, staining procedure) result in WSI with notably different visual presentations compared to training data. To ameliorate pre-analytic variances, approaches such as CycleGAN can be used to calibrate visual properties of images between sites, with the intent of improving DL classifier generalizability. In this work, we present a new approach termed Multi-Site Cross-Organ Calibration based Deep Learning (MuSClD) that employs WSIs of an off-target organ for calibration created at the same site as the on-target organ, based off the assumption that cross-organ slides are subjected to a common set of pre-analytical sources of variance. We demonstrate that by using an off-target organ from the test site to calibrate training data, the domain shift between training and testing data can be mitigated. Importantly, this strategy uniquely guards against potential data leakage introduced during calibration, wherein information only available in the testing data is imparted on the training data. We evaluate MuSClD in the context of the automated diagnosis of non-melanoma skin cancer (NMSC). Specifically, we evaluated MuSClD for identifying and distinguishing (a) basal cell carcinoma (BCC), (b) in-situ squamous cell carcinomas (SCC-In Situ), and (c) invasive squamous cell carcinomas (SCC-Invasive), using an Australian (training, n = 85) and a Swiss (held-out testing, n = 352) cohort. Our experiments reveal that MuSCID reduces the Wasserstein distances between sites in terms of color, contrast, and brightness metrics, without imparting noticeable artifacts to training data. The NMSC-subtyping performance is statistically improved as a result of MuSCID in terms of one-vs. rest AUC: BCC (0.92 vs 0.87, p = 0.01), SCC-In Situ (0.87 vs 0.73, p = 0.15) and SCC-Invasive (0.92 vs 0.82, p = 1e-5). Compared to baseline NMSC-subtyping with no calibration, the internal validation results of MuSClD (BCC (0.98), SCC-In Situ (0.92), and SCC-Invasive (0.97)) suggest that while domain shift indeed degrades classification performance, our on-target calibration using off-target tissue can safely compensate for pre-analytical variabilities, while improving the robustness of the model.     

Applying deep learning in digital breast tomosynthesis for automatic breast cancer detection: A review

The relatively recent reintroduction of deep learning has been a revolutionary force in the interpretation of diagnostic imaging studies. However, the technology used to acquire those images is undergoing a revolution itself at the very same time. Digital breast tomosynthesis (DBT) is one such technology, which has transformed the field of breast imaging. DBT, a form of three-dimensional mammography, is rapidly replacing the traditional two-dimensional mammograms. These parallel developments in both the acquisition and interpretation of breast images present a unique case study in how modern AI systems can be designed to adapt to new imaging methods. They also present a unique opportunity for co-development of both technologies that can better improve the validity of results and patient outcomes.In this review, we explore the ways in which deep learning can be best integrated into breast cancer screening workflows using DBT. We first explain the principles behind DBT itself and why it has become the gold standard in breast screening. We then survey the foundations of deep learning methods in diagnostic imaging, and review the current state of research into AI-based DBT interpretation. Finally, we present some of the limitations of integrating AI into clinical practice and the opportunities these present in this burgeoning field.     

A region-growing technique to improve multi-temporal DInSAR interferogram phase unwrapping performance

We present an efficient solution to mitigate phase unwrapping (PhU) errors that can affect a sequence of multi-temporal differential synthetic aperture radar (SAR) interferograms. To this aim, we propose a strategy that, starting from a properly chosen network of differential interferograms, complements PhU operations with an advanced multi-temporal region-growing (RG) procedure that exploits the space-time relationships among the computed interferograms. In particular, the proposed method implements an iterative procedure that, at each step, allows correcting a sequence of previously unwrapped interferograms at one selected pixel, namely candidate pixel, by exploiting the (unwrapped) phase values at its neighbouring ‘seed’ pixels (i.e. the ones already correctly unwrapped). Following their estimation, the unwrapped phases are then used to retrieve surface deformation products, such as mean deformation velocity maps and displacement time series, through (advanced) small baseline differential SAR interferometry (DInSAR) techniques. The effectiveness of the presented RG PhU algorithm is demonstrated by analysing a data set of SAR images acquired by the European Remote Sensing (ERS)-1/2 sensors over the megacity area of Istanbul, Turkey.     

Dynamic DIC by digital holography microscopy for enhancing phase-contrast visualization

Differential image contrast (DIC), through the numerical managing and manipulation of complex wavefronts obtained by digital holography (DH), is investigated. We name the approach Dynamical Differential Holographic Image Contrast (DDHIC). DDHIC dispenses from special optics and/or complex setup configurations with moveable components, as usually occurs in classical DIC, that is not well-suited for investigating objects experiencing dynamic evolution during the measurement. In fact, the technique presented here, is useful for floating samples since it allows, from a single recording, to set a posteriori the best conditions for DIC imaging in conjunction with the numerical focusing feature of DH. By DDHIC, the movies can be easily built-up to offering dynamic representation of phase-contrast along all directions, thus improving the visualization. Furthermore, the dynamic representation is useful for making the proper choice of other key parameters of DIC such as the amount of shear and the bias, with the aim to optimize the visualized phase-contrast imaging as favorite representation for bio-scientists. Investigation is performed on various biological samples.     

Importance of phase unwrapping for the reconstruction of microwave tomographic images

Microwave image reconstruction is typically based on a regularized least-square minimization of either the complex-valued field difference between recorded and modeled data or the logarithmic transformation of these field differences. Prior work has shown anecdotally that the latter outperforms the former in limited surveys of simulated and experimental phantom results. In this paper, we provide a theoretical explanation of these empirical findings by developing closed form solutions for the field and the inverted electromagnetic property parameters in one dimension which reveal the dependency of the estimated permittivity and conductivity on the absolute (unwrapped) phase of the measured signal at the receivers relative to the source transmission. The analysis predicts the poor performance of complex-valued field minimization as target size and/or frequency and electromagnetic contrast increase. Such poor performance is avoided by logarithmic transformation and preservation of absolute measured signal phase. Two-dimensional experiments based on both synthetic and clinical data are used to confirm these findings. Robustness of the logarithmic transformation to variation in the initial guess of the reconstructed target properties is also shown. The results are generalizable to three dimensions and indicate that the minimization form with logarithmic transformation offers image reconstruction performance characteristics that are much more desirable for medial microwave imaging applications relative to minimizing differences in complex-valued field quantities.     

Automated sedation outperforms manual administration of propofol and remifentanil in critically ill patients with deep sedation: a randomized phase II trial

Purpose

To compare automated administration of propofol and remifentanil guided by the Bispectral index (BIS) versus manual administration of short-acting drugs in critical care patients requiring deep sedation. The primary outcome was the percentage of BIS values between 40 and 60 (BIS_40–60).

Methods

This randomized controlled phase II trial in the intensive care unit (ICU) was conducted in adults with multiorgan failure. Thirty-one patients were assigned to receive sedation with propofol or remifentanil either by an automated or a manual system, both targeting BIS_40–60. Performance and feasibility of an automated administration were assessed.

Results

The study groups were well balanced in terms of demographic characteristics. Study duration averaged 18 [8–24] h in the automated group and 14 [9–21] h in the manual group (p = 0.81). Adequate sedation (BIS_40–60) was significantly more frequent in the automated group 77 [59–82] % than in the manual group 36 [22–56] %, with p = 0.001. Propofol consumption was reduced by a factor of 2 in the automated group with a median change of infusion rates of 39 ± 9 times per hour. In contrast, there were only 2 ± 1 propofol and 1 ± 1 remifentanil dose changes per hour in the manual group compared to 40 ± 9 for remifentanil in the automated group (p < 0.001). Vasopressors were more often discontinued or reduced in the automated group than in the manual control group (36 [6–40] vs. 12 [4–20] modifications, p = 0.03).

Conclusions

Continuous titration of propofol and remifentanil sedation with an automatic controller maintains deep sedation better than manual control in severely ill patients. It is associated with reduced sedative and vasopressor use.     

Spatio-temporal deep learning for automatic detection of intracranial vessel perforation in digital subtraction angiography during endovascular thrombectomy

Intracranial vessel perforation is a peri-procedural complication during endovascular therapy (EVT). Prompt recognition is important as its occurrence is strongly associated with unfavorable treatment outcomes. However, perforations can be hard to detect because they are rare, can be subtle, and the interventionalist is working under time pressure and focused on treatment of vessel occlusions. Automatic detection holds potential to improve rapid identification of intracranial vessel perforation. In this work, we present the first study on automated perforation detection and localization on X-ray digital subtraction angiography (DSA) image series. We adapt several state-of-the-art single-frame detectors and further propose temporal modules to learn the progressive dynamics of contrast extravasation. Application-tailored loss function and post-processing techniques are designed. We train and validate various automated methods using two national multi-center datasets (i.e., MR CLEAN Registry and MR CLEAN-NoIV Trial), and one international multi-trial dataset (i.e., the HERMES collaboration). With ten-fold cross-validation, the proposed methods achieve an area under the curve (AUC) of the receiver operating characteristic of 0.93 in terms of series level perforation classification. Perforation localization precision and recall reach 0.83 and 0.70 respectively. Furthermore, we demonstrate that the proposed automatic solutions perform at similar level as an expert radiologist.     

Influence of defocus on quantitative analysis of microscopic objects and individual cells with digital holography

Digital holography offers a unique method for studying microscopic objects using quantitative measurements of the optical phase delays of transmitted light. The optical phase may be integrated across the object to produce an optical volume measurement, a parameter related to dry mass by a simple scaling factor. While digital holography is useful for comparing the properties of microscopic objects, especially cells, we show here that quantitative comparisons of optical phase can be influenced by the focal plane of the measurement. Although holographic images can be refocused digitally using Fresnel propagation, ambiguity can result if this aspect is not carefully controlled. We demonstrate that microscopic objects can be accurately profiled by employing a digital refocusing method to analyze phase profiles of polystyrene microspheres and red blood cells.OCIS codes: (090.1995) Digital holography, (170.1530) Cell analysis, (120.5050) Phase measurement     

Measurement of the traction force of biological cells by digital holography

The traction force produced by biological cells has been visualized as distortions in flexible substrata. We have utilized quantitative phase microscopy by digital holography (DH-QPM) to study the wrinkling of a silicone rubber film by motile fibroblasts. Surface deformation and the cellular traction force have been measured from phase profiles in a direct and straightforward manner. DH-QPM is shown to provide highly efficient and versatile means for quantitatively analyzing cellular motility.     

指动脉皮瓣修复手指深度烧伤——附32例报告

目的:评价指动脉皮瓣修复手指深度烧伤的临床效果。方法:手指深度烧伤伴肌腱及指骨外露患者32例共36指,应用带指动脉、神经蒂顺行岛状皮瓣推进术或指固有动脉逆行岛状皮瓣转移术(包含吻合指神经背侧支指动脉逆行岛状皮瓣转移术)治疗,分别于术后3、6、12个月运用美国手外科学会总主动活动度(TAM)系统评定标准和按英国医学研究会(BMRC)标准,对患者供瓣区及被修复手指的感觉、运动、外观以及生活和工作质量进行评价。结果:术后半个月36指皮瓣全部存活,所有伤指运动功能恢复良好,无明显关节活动受限;长度良好,色泽正常,外观不臃肿,指腹饱满,质地柔软;指固有动脉顺行岛状皮瓣及吻合指固有神经背侧支指动脉逆行岛状皮瓣感觉恢复好;日常生活不受影响且恢复了工作。29例32指获完整随访,其中行带指动脉、神经蒂顺行岛状皮瓣推进术9指的术后3、6、12个月的平均综合评定均为优;指固有动脉逆行岛状皮瓣转移术15指术后3、6、12个月的平均综合评定分别为良、良、优,吻合指神经背侧支指动脉逆行岛状皮瓣转移术8指术后3、6、12个月的平均综合评定分别为良、优、优。结论:该类术式简单,可一次完成,皮瓣外形佳,是目前较为理想的治疗方法,使手指保持良好的功能与形态,供区功能和外形也较好。     

Fundamental precision bounds for three-dimensional optical localization microscopy using self-interference digital holography

Localization based microscopy using self-interference digital holography (SIDH) provides three-dimensional (3D) positional information about point sources with nanometer scale precision. To understand the performance limits of SIDH, here we calculate the theoretical limit to localization precision for SIDH when designed with two different configurations. One configuration creates the hologram using a plane wave and a spherical wave while the second configuration creates the hologram using two spherical waves. We further compare the calculated precision bounds to the 3D single molecule localization precision from different Point Spread Functions. SIDH results in almost constant localization precision in all three dimensions for a 20 µm thick depth of field. For high signal-to-background ratio (SBR), SIDH on average achieves better localization precision. For lower SBR values, the large size of the hologram on the detector becomes a problem, and PSF models perform better.     

Classification of small-scale hyperspectral images with multi-source deep transfer learning

ABSTRACT

Recently, many methods based on deep learning (DL) have been used for hyperspectral image (HSI) classification and achieved good performance. But such approaches often need numerous labelled training samples. This issue is aggravated when applying DL on small-scale HSIs. To alleviate this problem, transfer learning (TL) is introduced to HSI analysis by some existing works. Most of these works transfer knowledge from a single source domain to the target domain. However, the single source TL tends to learn specific knowledge instead of general knowledge. Moreover, since the samples are limited in one source domain, it only partially alleviates the shortage of labelled samples. To learn more general knowledge and further alleviate the issue of limited samples, we introduce the multi-source transfer learning strategy to classify HSIs. Specifically, a framework named multi-source deep transfer learning (MS-DTL) is proposed. This framework consists of a multi-source compatible model and a customized loss function. We perform experiments by comparing the proposed method with the baseline methods on the well-known hyperspectral datasets. The results show that the proposed MS-DTL performs better than the benchmarks on the classification tasks of the small-scale HSIs. Thanks to the strategy of TL, the proposed network is also time-saving.