KFPredict: An ensemble learning prediction framework for diabetes based on fusion of key features

Computer Methods and Programs in Biomedicine, Год журнала: 2023, Номер 231, С. 107378 - 107378

Опубликована: Янв. 26, 2023

Язык: Английский

---

Vision transformer promotes cancer diagnosis: A comprehensive review

Expert Systems with Applications, Год журнала: 2024, Номер 252, С. 124113 - 124113

Опубликована: Май 1, 2024

Язык: Английский

---

Deep Learning to Simulate Contrast-Enhanced MRI for Evaluating Suspected Prostate Cancer

Radiology, Год журнала: 2025, Номер 314(1)

Опубликована: Янв. 1, 2025

Background Multiparametric MRI, including contrast-enhanced sequences, is recommended for evaluating suspected prostate cancer, but concerns have been raised regarding potential contrast agent accumulation and toxicity. Purpose To evaluate the feasibility of generating simulated MRI from noncontrast sequences using deep learning to explore their value assessing clinically significant cancer Prostate Imaging Reporting Data System (PI-RADS) version 2.1. Materials Methods Male patients with who underwent multiparametric were retrospectively included three centers April 2020 2023. A model (pix2pix algorithm) was trained synthesize scans four (T1-weighted imaging, T2-weighted diffusion-weighted apparent diffusion coefficient maps) then tested on an internal two external datasets. The reference standard training second postcontrast phase dynamic sequence. Similarity between acquired images evaluated multiscale structural similarity index. Three radiologists independently scored either or PI-RADS, 2.1; agreement assessed Cohen κ. Results total 567 male (mean age, 66 years ± 11 [SD]) divided into a test set (n = 244), 104), 1 143), 2 76). Simulated demonstrated high (multiscale index: 0.82, 0.71, 0.69 set, 1, 2, respectively) excellent reader PI-RADS scores (Cohen κ, 0.96; 95% CI: 0.94, 0.98). When imaging added biparametric 34 323 (10.5%) upgraded 4 3. Conclusion It feasible generate learning. exhibited in based © RSNA, 2025 Supplemental material available this article. See also editorial by Neji Goh issue.

Язык: Английский

---

VGG16-based intelligent image analysis in the pathological diagnosis of IgA nephropathy

Journal of Radiation Research and Applied Sciences, Год журнала: 2023, Номер 16(3), С. 100626 - 100626

Опубликована: Июль 12, 2023

IgA nephropathy (IgAN) is the most common primary glomerular disease worldwide, with heterogeneous clinical and pathological manifestations, a cause of end-stage renal disease. Early detection effective intervention measures are essential for improving outcome IgAN. Machine learning methods can make analysis, early diagnosis, prognosis prediction IgAN more automated accurate. This article discusses application machine in IgAN, from optimizing diagnosis to discovering non-invasive specific biomarkers, predicting progression, evaluating prognosis. It key reducing incidence rate mortality by relying on intelligent image analysis VGG16 accurate enable take prevention treatment measures. A total 452 cases kidney admitted nephrology department our hospital January 2018 February 2023 were selected, it was ruled out that could not be made due small number samples submitted puncture; After excluding suspected biopsy pathology patients who did undergo immunofluorescence examination, 135 confirmed subjected analysis. The internationally recognized 5-level semi-quantitative method used evaluation, traditional processing selected segment extract fluorescence deposition areas. Transform input into color space generate binary using adaptive threshold two feature dimensions brightness. Then, regions separated merged obtain independent sedimentary regions. add BN layers SE visual attention fully sensitive features high inter-class similarity significant intra-class differences classification task. contour, area, average brightness each region calculated, automatic computer recognition intensity shape obtained improve accuracy classification. artificial intelligence based achieve interpretation results higher coincidence compared diagnostic doctors. reaches 88.9%, IgG 85.8%, IgM 83.8%, C3 88.6%. Therefore, assist doctors interpreting immunofluorescence. By utilizing network technologies change workflow, work efficiency doctors, reduce misdiagnosis caused fatigue during film reading, objective.

Язык: Английский

---

Artificial general intelligence for radiation oncology

Meta-Radiology, Год журнала: 2023, Номер 1(3), С. 100045 - 100045

Опубликована: Ноя. 1, 2023

The emergence of artificial general intelligence (AGI) is transforming radiation oncology. As prominent vanguards AGI, large language models (LLMs) such as GPT-4 and PaLM 2 can process extensive texts vision (LVMs) the Segment Anything Model (SAM) imaging data to enhance efficiency precision therapy. This paper explores full-spectrum applications AGI across oncology including initial consultation, simulation, treatment planning, delivery, verification, patient follow-up. fusion with LLMs also creates powerful multimodal that elucidate nuanced clinical patterns. Together, promises catalyze a shift towards data-driven, personalized However, these should complement human expertise care. provides an overview how transform elevate standard care in oncology, key insight being AGI's ability exploit at scale.

Язык: Английский

---

Pelvic bone tumor segmentation fusion algorithm based on fully convolutional neural network and conditional random field

Journal of bone oncology, Год журнала: 2024, Номер 45, С. 100593 - 100593

Опубликована: Фев. 28, 2024

Pelvic bone tumors represent a harmful orthopedic condition, encompassing both benign and malignant forms. Addressing the issue of limited accuracy in current machine learning algorithms for tumor image segmentation, we have developed an enhanced segmentation algorithm. This algorithm is built upon improved full convolutional neural network, incorporating fully network (FCNN-4s) conditional random field (CRF) to achieve more precise segmentation. The was employed conduct initial on preprocessed images. Following each layer, batch normalization layers were introduced expedite training convergence enhance trained model. Subsequently, connected integrated fine-tune results, refining boundaries pelvic achieving high-quality experimental outcomes demonstrate significant enhancement stability when compared conventional achieves average Dice coefficient 93.31%, indicating superior performance real-time operations. In contrast algorithm, presented this paper boasts intricate structure, proficiently addressing issues over-segmentation under-segmentation model exhibits performance, robust stability, capable heightened accuracy.

Язык: Английский

---

Comparative study of deep learning models on the images of biopsy specimens for diagnosis of lung cancer treatment

Journal of Radiation Research and Applied Sciences, Год журнала: 2023, Номер 16(2), С. 100555 - 100555

Опубликована: Март 8, 2023

To compare the application value of different traditional deep learning models in diagnosing and classifying lung cancer. According to biopsy samples our hospital from January 2018 November 2022, 37 patients treated this department were selected as study subjects. Nonsmall small cell cancer specimens obtained stained. Two experienced pathologists diagnosed specimens. Multiple in-depth methods used distinguish between noncancer biopsies. In study, we compared diagnosis Classification. The tested several popular CNN architectures based on image block classification: AlexNet, VGG, ResNet, SqueezeNet, comparing two types training schemes: scratch fine-tuning entire pretrained network. AUC model is more reasonable (0. 8808–0. 9121). Except for Resnet-50, higher than that whole Deep analysis can accelerate detection speed section images (WSI) maintain a similar rate with pathologists.

Язык: Английский

---

Deep learning-based magnetic resonance imaging of the spine in the diagnosis and physiological evaluation of spinal metastases

Journal of bone oncology, Год журнала: 2023, Номер 40, С. 100483 - 100483

Опубликована: Май 9, 2023

Spinal metastasis accounts for 70% of the bone metastases tumors, so how to diagnose and predict spinal in time through effective methods is very important physiological evaluation therapy patients.MRI scans 941 patients with from affiliated hospital Guilin Medical University were collected, analyzed, preprocessed, data submitted a deep learning model designed our convolutional neural network. We also used Softmax classifier classify results compared them actual judge accuracy model.Our research showed that practical method could effectively metastases. The was up 96.45%, which be metastases.The obtained final experiment can capture focal signs more accurately disease time, has good application prospect.

Язык: Английский

---

Mask-guided generative adversarial network for MRI-based CT synthesis

Knowledge-Based Systems, Год журнала: 2024, Номер 295, С. 111799 - 111799

Опубликована: Апрель 16, 2024

Язык: Английский

---

Boundary information‐guided adversarial diffusion model for efficient unsupervised synthetic CT generation

Medical Physics, Год журнала: 2025, Номер unknown

Опубликована: Фев. 28, 2025

Abstract Background The absence of tissue electron density information derived from greyscale Hounsfield units (HUs) in magnetic resonance imaging (MRI) limits its further clinical application radiotherapy (RT). use synthetic computed tomography (sCT) with MRI simplifies RT treatment and improves positioning accuracy by eliminating the need for (CT) simulation radiation dose error‐prone image registration. Although CycleGAN variants can obtain verisimilar sCT through unsupervised learning, ensuring perfect structural consistency synthesized images this approach remains challenging, thus limiting quality diversity a given application. Purpose purpose work is to develop novel boundary information‐guided adversarial diffusion model, called RadADM, aim enhancing performance regard unpaired MR‐to‐CT translation MR‐only RT. Methods In order explicitly guide feature learning proposed RadADM mask incorporated as guidance anatomy compensation during generation simulated MR images. addition, cycle‐consistent module incorporates projections featuring coupled diffusive non‐diffusive architecture used facilitate training on MR‐CT datasets, enabling accurate efficient between source target domain To validate we conducted comprehensive quantitative qualitative comparison other state‐of‐the‐art methods, including CycleGAN, CycleSlimulationGAN, CUT, Fixed Learned Self‐Similarity (F‐LseSim), SynDiff. Results We evaluated demonstrated that outperforms comparative approaches high‐quality pelvic captures local features, achieves smaller errors mean absolute error (MAE): 62.95 ± 23.15 root square (RMSE): 135.46 23.89 higher similarities peak signal‐to‐noise ratio (PSNR): 24.70 0.52, similarity index (SSIM): 0.8673 0.01. For region soft‐tissue, PSNR SSIM were 33.99 1.09 0.931 0.01, bone, 35.79 0.87 0.993 0.04. Conclusions Extensive experiments datasets demonstrate effectiveness robustness our terms synthesizing at anatomical level. Our found offer valuable promising direction adaptive cancer.

Язык: Английский

---