Intraoperative Glioma Grading Using Neural Architecture Search and Multi-Modal Imaging

A review of AutoML optimization techniques for medical image applications

Automatic analysis of medical images using machine learning techniques has gained significant importance over the years. A large number of approaches have been proposed for solving different medical image analysis tasks using machine learning and deep learning approaches. These approaches are quite effective thanks to their ability to analyze large volume of medical imaging data. Moreover, they can also identify patterns that may be difficult for human experts to detect. Manually designing and tuning the parameters of these algorithms is a challenging and time-consuming task. Furthermore, designing a generalized model that can handle different imaging modalities is difficult, as each modality has specific characteristics. To solve these problems and automate the whole pipeline of different medical image analysis tasks, numerous Automatic Machine Learning (AutoML) techniques have been proposed. These techniques include Hyper-parameter Optimization (HPO), Neural Architecture Search (NAS), and Automatic Data Augmentation (ADA). This study provides an overview of several AutoML-based approaches for different medical imaging tasks in terms of optimization search strategies. The usage of optimization techniques (evolutionary, gradient-based, Bayesian optimization, etc.) is of significant importance for these AutoML approaches. We comprehensively reviewed existing AutoML approaches, categorized them, and performed a detailed analysis of different proposed approaches. Furthermore, current challenges and possible future research directions are also discussed.

Illuminating the future of precision cancer surgery with fluorescence imaging and artificial intelligence convergence

Real-time and accurate guidance for tumor resection has long been anticipated by surgeons. In the past decade, the flourishing material science has made impressive progress in near-infrared fluorophores that may fulfill this purpose. Fluorescence imaging-guided surgery shows great promise for clinical application and has undergone widespread evaluations, though it still requires continuous improvements to transition this technique from bench to bedside. Concurrently, the rapid progress of artificial intelligence (AI) has revolutionized medicine, aiding in the screening, diagnosis, and treatment of human doctors. Incorporating AI helps enhance fluorescence imaging and is poised to bring major innovations to surgical guidance, thereby realizing precision cancer surgery. This review provides an overview of the principles and clinical evaluations of fluorescence-guided surgery. Furthermore, recent endeavors to synergize AI with fluorescence imaging were presented, and the benefits of this interdisciplinary convergence were discussed. Finally, several implementation strategies to overcome technical hurdles were proposed to encourage and inspire future research to expedite the clinical application of these revolutionary technologies.

DIPO: Differentiable Parallel Operation Blocks for Surgical Neural Architecture Search

Deep learning has been used across a large number of computer vision tasks, however designing the network architectures for each task is time consuming. Neural Architecture Search (NAS) promises to automatically build neural networks, optimised for the given task and dataset. However, most NAS methods are constrained to a specific macro-architecture design which makes it hard to apply to different tasks (classification, detection, segmentation). Following the work in Differentiable NAS (DNAS), we present a simple and efficient NAS method, Differentiable Parallel Operation (DIPO), that constructs a local search space in the form of a DIPO block, and can easily be applied to any convolutional network by injecting it in-place of the convolutions. The DIPO block's internal architecture and parameters are automatically optimised end-to-end for each task. We demonstrate the flexibility of our approach by applying DIPO to 4 model architectures (U-Net, HRNET, KAPAO and YOLOX) across different surgical tasks (surgical scene segmentation, surgical instrument detection, and surgical instrument pose estimation) and evaluated across 5 datasets. Results show significant improvements in surgical scene segmentation (+10.5% in CholecSeg8K, +13.2% in CaDIS), instrument detection (+1.5% in ROBUST-MIS, +5.3% in RoboKP), and instrument pose estimation (+9.8% in RoboKP).

PP-NAS: Searching for Plug-and-Play Blocks on Convolutional Neural Networks

Multiscale features are of great importance in modern convolutional neural networks, showing consistent performance gains on numerous vision tasks. Therefore, many plug-and-play blocks are introduced to upgrade existing convolutional neural networks for stronger multiscale representation ability. However, the design of plug-and-play blocks is getting more and more complex, and these manually designed blocks are not optimal. In this work, we propose PP-NAS to develop plug-and-play blocks based on neural architecture search (NAS). Specifically, we design a new search space PPConv and develop a search algorithm consisting of one-level optimization, zero-one loss, and connection existence loss. PP-NAS minimizes the optimization gap between super-net and subarchitectures and can achieve good performance even without retraining. Extensive experiments on image classification, object detection, and semantic segmentation verify the superiority of PP-NAS over state-of-the-art CNNs (e.g., ResNet, ResNeXt, and Res2Net). Our code is available at https://github.com/ainieli/PP-NAS.

NIR-II light in clinical oncology: opportunities and challenges

Novel strategies utilizing light in the second near-infrared region (NIR-II; 900-1,880 nm wavelengths) offer the potential to visualize and treat solid tumours with enhanced precision. Over the past few decades, numerous techniques leveraging NIR-II light have been developed with the aim of precisely eliminating tumours while maximally preserving organ function. During cancer surgery, NIR-II optical imaging enables the visualization of clinically occult lesions and surrounding vital structures with increased sensitivity and resolution, thereby enhancing surgical quality and improving patient prognosis. Furthermore, the use of NIR-II light promises to improve cancer phototherapy by enabling the selective delivery of increased therapeutic energy to tissues at greater depths. Initial clinical studies of NIR-II-based imaging and phototherapy have indicated impressive potential to decrease cancer recurrence, reduce complications and prolong survival. Despite the encouraging results achieved, clinical translation of innovative NIR-II techniques remains challenging and inefficient; multidisciplinary cooperation is necessary to bridge the gap between preclinical research and clinical practice, and thus accelerate the translation of technical advances into clinical benefits. In this Review, we summarize the available clinical data on NIR-II-based imaging and phototherapy, demonstrating the feasibility and utility of integrating these technologies into the treatment of cancer. We also introduce emerging NIR-II-based approaches with substantial potential to further enhance patient outcomes, while also highlighting the challenges associated with imminent clinical studies of these modalities.

Cellular nucleus image-based smarter microscope system for single cell analysis

Cell imaging technology is undoubtedly a powerful tool for studying single-cell heterogeneity due to its non-invasive and visual advantages. It covers microscope hardware, software, and image analysis techniques, which are hindered by low throughput owing to abundant hands-on time and expertise. Herein, a cellular nucleus image-based smarter microscope system for single-cell analysis is reported to achieve high-throughput analysis and high-content detection of cells. By combining the hardware of an automatic fluorescence microscope and multi-object recognition/acquisition software, we have achieved more advanced process automation with the assistance of Robotic Process Automation (RPA), which realizes a high-throughput collection of single-cell images. Automated acquisition of single-cell images has benefits beyond ease and throughout and can lead to uniform standard and higher quality images. We further constructed a single-cell image database-based convolutional neural network (Efficient Convolutional Neural Network, E-CNN) exceeding 20618 single-cell nucleus images. Computational analysis of large and complex data sets enhances the content and efficiency of single-cell analysis with the assistance of Artificial Intelligence (AI), which breaks through the super-resolution microscope's hardware limitation, such as specialized light sources with specific wavelengths, advanced optical components, and high-performance graphics cards. Our system can identify single-cell nucleus images that cannot be artificially distinguished with an accuracy of 95.3%. Overall, we build an ordinary microscope into a high-throughput analysis and high-content smarter microscope system, making it a candidate tool for Imaging cytology.

Real-time classification of tumour and non-tumour tissue in colorectal cancer using diffuse reflectance spectroscopy and neural networks to aid margin assessment

BACKGROUND

Colorectal cancer is the third most commonly diagnosed malignancy and the second leading cause of mortality worldwide. A positive resection margin following surgery for colorectal cancer is linked with higher rates of local recurrence and poorer survival. We investigated diffuse reflectance spectroscopy (DRS) to distinguish tumour and non-tumour tissue in ex vivo colorectal specimens, to aid margin assessment and provide augmented visual maps to the surgeon in real-time.

METHODS

Patients undergoing elective colorectal cancer resection surgery at a London-based hospital were prospectively recruited. A hand-held DRS probe was used on the surface of freshly resected ex vivo colorectal tissue. Spectral data was acquired for tumour and non-tumour tissue. Binary classification was achieved using conventional machine learning classifiers and a convolutional neural network (CNN), which were evaluated in terms of sensitivity, specificity, accuracy and the area under the curve.

RESULTS

A total of 7692 mean spectra were obtained for tumour and non-tumour colorectal tissue. The CNN-based classifier was the best performing machine learning algorithm, when compared to contrastive approaches, for differentiating tumour and non-tumour colorectal tissue, with an overall diagnostic accuracy of 90.8% and area under the curve of 96.8%. Live on-screen classification of tissue type was achieved using a graduated colourmap.

CONCLUSION

A high diagnostic accuracy for a DRS probe and tracking system to differentiate ex vivo tumour and non-tumour colorectal tissue in real-time with on-screen visual feedback was highlighted by this study. Further in vivo studies are needed to ensure integration into a surgical workflow.

Rapid intraoperative multi-molecular diagnosis of glioma with ultrasound radio frequency signals and deep learning

BACKGROUND

Molecular diagnosis is crucial for biomarker-assisted glioma resection and management. However, some limitations of current molecular diagnostic techniques prevent their widespread use intraoperatively. With the unique advantages of ultrasound, this study developed a rapid intraoperative molecular diagnostic method based on ultrasound radio-frequency signals.

METHODS

We built a brain tumor ultrasound bank with 169 cases enrolled since July 2020, of which 43483 RF signal patches from 67 cases with a pathological diagnosis of glioma were a retrospective cohort for model training and validation. IDH1 and TERT promoter (TERTp) mutations and 1p/19q co-deletion were detected by next-generation sequencing. We designed a spatial-temporal integration model (STIM) to diagnose the three molecular biomarkers, thus establishing a rapid intraoperative molecular diagnostic system for glioma, and further analysed its consistency with the fifth edition of the WHO Classification of Tumors of the Central Nervous System (WHO CNS5). We tested STIM in 16-case prospective cohorts, which contained a total of 10384 RF signal patches. Two other RF-based classical models were used for comparison. Further, we included 20 cases additional prospective data for robustness test (ClinicalTrials.govNCT05656053).

FINDINGS

In the retrospective cohort, STIM achieved a mean accuracy and AUC of 0.9190 and 0.9650 (95% CI, 0.94-0.99) respectively for the three molecular biomarkers, with a total time of 3 s and a 96% match to WHO CNS5. In the prospective cohort, the diagnostic accuracy of STIM is 0.85 ± 0.04 (mean ± SD) for IDH1, 0.84 ± 0.05 for TERTp, and 0.88 ± 0.04 for 1p/19q. The AUC is 0.89 ± 0.02 (95% CI, 0.84-0.94) for IDH1, 0.80 ± 0.04 (95% CI, 0.71-0.89) for TERTp, and 0.85 ± 0.06 (95% CI, 0.73-0.98) for 1p/19q. Compared to the second best available method based on RF signal, the diagnostic accuracy of STIM is improved by 16.70% and the AUC is improved by 19.23% on average.

INTERPRETATION

STIM is a rapid, cost-effective, and easy-to-manipulate AI method to perform real-time intraoperative molecular diagnosis. In the future, it may help neurosurgeons designate personalized surgical plans and predict survival outcomes.

FUNDING

A full list of funding bodies that contributed to this study can be found in the Acknowledgements section.

Performance of deep learning algorithms to distinguish high-grade glioma from low-grade glioma: A systematic review and meta-analysis

This study aims to evaluate deep learning (DL) performance in differentiating low- and high-grade glioma. Search online database for studies continuously published from 1st January 2015 until 16th August 2022. The random-effects model was used for synthesis, based on pooled sensitivity (SE), specificity (SP), and area under the curve (AUC). Heterogeneity was estimated using the Higgins inconsistency index (I²). 33 were ultimately included in the meta-analysis. The overall pooled SE and SP were 94% and 93%, with an AUC of 0.98. There was great heterogeneity in this field. Our evidence-based study shows DL achieves high accuracy in glioma grading. Subgroup analysis reveals several limitations in this field: 1) Diagnostic trials require standard method for data merging for AI; 2) small sample size; 3) poor-quality image preprocessing; 4) not standard algorithm development; 5) not standard data report; 6) different definition of HGG and LGG; and 7) poor extrapolation.