The use of machine learning has seen extraordinary growth since the development of deep learning techniques, notably the deep artificial neural network. Deep learning methodology excels in addressing complicated problems such as image classification, object detection, and natural language processing. A key feature of these networks is the capability to extract useful patterns from vast quantities of complex data, including images. As many branches of healthcare revolves around the generation, processing, and analysis of images, these techniques have become increasingly commonplace. This is especially true for radiotherapy, which relies on the use of anatomical and functional images from a range of imaging modalities, such as Computed Tomography (CT). The aim of this review is to provide an understanding of deep learning methodologies, including neural network types and structure, as well as linking these general concepts to medical CT image processing for radiotherapy. Specifically, it focusses on the stages of enhancement and analysis, incorporating image denoising, super-resolution, generation, registration, and segmentation, supported by examples of recent literature.

Artificial intelligence (AI) and combinatorial optimization drive applications across science and industry, but their increasing energy demands challenge the sustainability of digital computing. Most unconventional computing systems^1-7 target either AI or optimization workloads and rely on frequent, energy-intensive digital conversions, limiting efficiency. These systems also face application-hardware mismatches, whether handling memory-bottlenecked neural models, mapping real-world optimization problems or contending with inherent analog noise. Here we introduce an analog optical computer (AOC) that combines analog electronics and three-dimensional optics to accelerate AI inference and combinatorial optimization in a single platform. This dual-domain capability is enabled by a rapid fixed-point search, which avoids digital conversions and enhances noise robustness. With this fixed-point abstraction, the AOC implements emerging compute-bound neural models with recursive reasoning potential and realizes an advanced gradient-descent approach for expressive optimization. We demonstrate the benefits of co-designing the hardware and abstraction, echoing the co-evolution of digital accelerators and deep learning models, through four case studies: image classification, nonlinear regression, medical image reconstruction and financial transaction settlement. Built with scalable, consumer-grade technologies, the AOC paves a promising path for faster and sustainable computing. Its native support for iterative, compute-intensive models offers a scalable analog platform for fostering future innovation in AI and optimization.

The primary aim of this research was to create and rigorously assess a deep learning radiomics (DLR) framework utilizing magnetic resonance imaging (MRI) to forecast the histological differentiation grades of oropharyngeal cancer. This retrospective analysis encompassed 122 patients diagnosed with oropharyngeal cancer across three medical institutions in China. The participants were divided at random into two groups: a training cohort comprising 85 individuals and a test cohort of 37. Radiomics features derived from MRI scans, along with deep learning (DL) features, were meticulously extracted and carefully refined. These two sets of features were then integrated to build the DLR model, designed to assess the histological differentiation of oropharyngeal cancer. The model's predictive efficacy was gaged through the area under the receiver operating characteristic curve (AUC) and decision curve analysis (DCA). The DLR model demonstrated impressive performance, achieving strong AUC scores of 0.871 on the training cohort and 0.803 on the test cohort, outperforming both the standalone radiomics and DL models. Additionally, the DCA curve highlighted the significance of the DLR model in forecasting the histological differentiation of oropharyngeal cancer. The MRI-based DLR model demonstrated high predictive ability for histological differentiation of oropharyngeal cancer, which might be important for accurate preoperative diagnosis and clinical decision-making.

Non-small cell lung cancer (NSCLC) is a complex disease characterized by diverse clinical, genetic, and histopathologic traits, necessitating personalized treatment approaches. While numerous biomarkers have been introduced for NSCLC prognostication, no single source of information can provide a comprehensive understanding of the disease. However, integrating biomarkers from multiple sources may offer a holistic view of the disease, enabling more accurate predictions. In this study, we present MetaPredictomics, a framework that integrates clinicopathologic data with PET/CT radiomics from the primary tumor and presumed healthy organs (referred to as "organomics") to predict postsurgical recurrence. A fully automated deep learning-based segmentation model was employed to delineate 19 affected (whole lung and the affected lobe) and presumed healthy organs from CT images of the presurgical PET/CT scans of 145 NSCLC patients sourced from a publicly available data set. Using PyRadiomics, 214 features (107 from CT, 107 from PET) were extracted from the gross tumor volume (GTV) and each segmented organ. In addition, a clinicopathologic feature set was constructed, incorporating clinical characteristics, histopathologic data, gene mutation status, conventional PET imaging biomarkers, and patients' treatment history. GTV Radiomics, each of the organomics, and the clinicopathologic feature sets were each fed to a time-to-event prediction machine, based on glmboost, to establish first-level models. The risk scores obtained from the first-level models were then used as inputs for meta models developed using a stacked ensemble approach. Questing optimized performance, we assessed meta models established upon all combinations of first-level models with concordance index (C-index) ≥0.6. The performance of all the models was evaluated using the average C-index across a unique 3-fold cross-validation scheme for fair comparison. The clinicopathologic model outperformed other first-level models with a C-index of 0.67, followed closely by GTV radiomics model with C-index of 0.65. Among the organomics models, whole-lung and aorta models achieved top performance with a C-index of 0.65, while 12 organomics models achieved C-indices of ≥0.6. Meta models significantly outperformed the first-level models with the top 100 achieving C-indices between 0.703 and 0.731. The clinicopathologic, whole lung, esophagus, pancreas, and GTV models were the most frequently present models in the top 100 meta models with frequencies of 98, 71, 69, 62, and 61, respectively. In this study, we highlighted the value of maximizing the use of medical imaging for NSCLC recurrence prognostication by incorporating data from various organs, rather than focusing solely on the tumor and its immediate surroundings. This multisource integration proved particularly beneficial in the meta models, where combining clinicopathologic data with tumor radiomics and organomics models significantly enhanced recurrence prediction.

not requested for this type of paper.

High mammographic density (MD) and excess weight are both associated with increased risk of breast cancer. Classically defined percentage density measures tend to increase with reduced weight due to disproportionate loss of breast fat, however the effect of weight loss on artificial intelligence-based density scores is unknown. We investigated an artificial intelligence-based density method, reporting density changes in 46 women enrolled in a weight-loss study in a family history breast cancer clinic, using a volumetric density method as a comparison.&#xD;&#xD;Methods: We analysed data from women who had weight recorded and mammograms taken at the start and end of the 12-month weight intervention study. MD was assessed at both time points using a deep learning model trained on expert estimates of percent density called pVAS, and the volumetric density software VolparaTM.&#xD;&#xD;Results: Mean (standard deviation) weight of participants at the start and end of the study was 86.0 (12.2) and 82.5 (13.8) respectively; mean (standard deviation) pVAS scores were 35.8 (13.0) and 36.3 (12.4), and Volpara volumetric percent density scores were 7.05 (4.4) and 7.6 (4.4).The Spearman rank correlation between reduction in weight and change in density was 0.17 (-0.13 to 0.43, p=0.27) for pVAS and 0.59 (0.36 to 0.75, p<0.001) for Volpara volumetric percent density.&#xD;&#xD;Conclusion: pVAS percentage density measurements were not significantly affected by change in weight. Percent density measured with Volpara increased as weight decreased, driven by changes in fat volume.&#xD.

Objective&#xD;Low-dose interior tomography integrates low-dose CT (LDCT) with region-of-interest (ROI) imaging which finds wide application in radiation dose reduction and high-resolution imaging. However, the combined effects of noise and data truncation pose great challenges for accurate tomographic reconstruction. This study aims to develop a novel reconstruction framework that achieves high-quality ROI reconstruction and efficient extension of recoverable region to provide innovative solutions to address coupled ill-posed problems.&#xD;Approach&#xD;We conducted a comprehensive analysis of projection data composition and angular sampling patterns in low-dose interior tomography. Based on this analysis, we proposed two novel deep learning-based reconstruction pipelines: (1) Deep Projection Extraction-based Reconstruction (DPER) that focuses on ROI reconstruction by disentangling and extracting noise and background projection contributions using a dual-domain deep neural network; and (2) DPER with Progressive extension (DPER-Pro) that enhances DPER by a progressive "coarse-to-fine" strategy for missing data compensation, enabling simultaneous ROI reconstruction and extension of recoverable regions. The proposed methods were rigorously evaluated through extensive experiments on simulated torso datasets and real CT scans of a torso phantom.&#xD;Main Results&#xD;The experimental results demonstrated that DPER effectively handles the coupled ill-posed problem and achieves high-quality ROI reconstructions by accurately extracting noise and background projections. DPER-Pro extends the recoverable region while preserving ROI image quality by leveraging disentangled projection components and angular sampling patterns. Both methods outperform competing approaches in reconstructing reliable structures, enhancing generalization, and mitigating noise and truncation artifacts.&#xD;Significance&#xD;This work presents a novel decoupled deep learning framework for low-dose interior tomography that provides a robust and effective solution to the challenges posed by noise and truncated projections. The proposed methods significantly improve ROI reconstruction quality while efficiently recovering structural information in exterior regions, offering a promising pathway for advancing low-dose ROI imaging across a wide range of applications.&#xD.

We propose a data-driven method for approximating real-valued functions on smooth manifolds, building on the Diffusion Maps framework under the manifold hypothesis. Given pointwise evaluations of a function, the method constructs a smooth extension to the ambient space by exploiting diffusion geometry and its connection to the heat equation and the Laplace-Beltrami operator. To address the computational challenges of high-dimensional data, we introduce a dimensionality reduction strategy based on the low-rank structure of the distance matrix, revealed via singular value decomposition (SVD). In addition, we develop an online updating mechanism that enables efficient incorporation of new data, thereby improving scalability and reducing computational cost. Numerical experiments, including applications to sparse CT reconstruction, demonstrate that the proposed methodology outperforms classical feedforward neural networks and interpolation methods in terms of both accuracy and efficiency.

MTV is increasingly recognized as an accurate estimate of disease burden, which has prognostic value, but its implementation has been hindered by the time-consuming need for manual segmentation of images. Automated quantitation using AI-driven approaches is promising. AI-driven automated quantification significantly reduces labor-intensive manual segmentation, improving consistency, reproducibility, and feasibility for routine clinical practice. AI-enhanced radiomics provides comprehensive characterization of tumor biology, capturing intratumoral and intertumoral heterogeneity beyond what conventional volumetric metrics alone offer, supporting improved patient stratification and therapy planning. AI-driven segmentation of normal organs improves radioligand therapy planning by enabling accurate dose predictions and comprehensive organ-based radiomics analysis, further refining personalized patient management.

To develop and validate a non-invasive deep learning model that integrates voxel-level radiomics with multi-sequence MRI to predict microsatellite instability (MSI) status in patients with endometrial carcinoma (EC). This two-center retrospective study included 375 patients with pathologically confirmed EC from two medical centers. Patients underwent preoperative multiparametric MRI (T2WI, DWI, CE-T1WI), and MSI status was determined by immunohistochemistry. Tumor regions were manually segmented, and voxel-level radiomics features were extracted following IBSI guidelines. A dual-channel 3D deep neural network based on the Vision-Mamba architecture was constructed to jointly process voxel-wise radiomics feature maps and MR images. The model was trained and internally validated on cohorts from Center I and tested on an external cohort from Center II. Performance was compared with Vision Transformer, 3D-ResNet, and traditional radiomics models. Interpretability was assessed with feature importance ranking and SHAP value visualization. The Vision-Mamba model achieved strong predictive performance across all datasets. In the external test cohort, it yielded an AUC of 0.866, accuracy of 0.875, sensitivity of 0.833, and specificity of 0.900, outperforming other models. Integrating voxel-level radiomics features with MRI enabled the model to better capture both local and global tumor heterogeneity compared to traditional approaches. Interpretability analysis identified glszm_SizeZoneNonUniformityNormalized, ngtdm_Busyness, and glcm_Correlation as top features, with SHAP analysis revealing that tumor parenchyma, regions of enhancement, and diffusion restriction were pivotal for MSI prediction. The proposed voxel-level radiomics and deep learning model provides a robust, non-invasive tool for predicting MSI status in endometrial carcinoma, potentially supporting personalized treatment decision-making.