Deep transformer-based personalized dosimetry from SPECT/CT images: a hybrid approach for [177Lu]Lu-DOTATATE radiopharmaceutical therapy

European Journal of Nuclear Medicine and Molecular Imaging, Год журнала: 2024, Номер 51(6), С. 1516 - 1529

Опубликована: Янв. 25, 2024

Abstract Purpose Accurate dosimetry is critical for ensuring the safety and efficacy of radiopharmaceutical therapies. In current clinical practice, MIRD formalisms are widely employed. However, with rapid advancement deep learning (DL) algorithms, there has been an increasing interest in leveraging calculation speed automation capabilities different tasks. We aimed to develop a hybrid transformer-based model that incorporates multiple voxel S -value (MSV) approach voxel-level [ 177 Lu]Lu-DOTATATE therapy. The goal was enhance performance achieve accuracy levels closely aligned Monte Carlo (MC) simulations, considered as standard reference. extended our analysis include (SSV MSV), thereby conducting comprehensive study. Methods used dataset consisting 22 patients undergoing up 4 cycles MC simulations were generate reference absorbed dose maps. addition, formalism approaches, namely, single (SSV) MSV techniques, performed. A UNEt TRansformer (UNETR) DL architecture trained using five-fold cross-validation MC-based Co-registered CT images fed into network input, whereas difference between (MC-MSV) set output. results then integrated revive Finally, maps generated by MSV, SSV, quantitatively compared at both level organ (organs risk lesions). Results showed slightly better (voxel relative absolute error (RAE) = 5.28 ± 1.32) RAE 5.54 1.4) outperformed SSV 7.8 3.02). Gamma pass rates 99.0 1.2%, 98.8 1.3%, 98.7 1.52% DL, respectively. computational time highest (~2 days single-bed SPECT study) DL-based other approaches terms efficiency (3 s SPECT). Organ-wise percent errors 1.44 3.05%, 1.18 2.65%, 1.15 2.5% respectively, lesion-absorbed doses. Conclusion developed fast accurate map generation, outperforming specifically heterogenous regions. achieved close gold potential implementation use on large-scale datasets.

Язык: Английский

---

Explainable AI in Diagnostic Radiology for Neurological Disorders: A Systematic Review, and What Doctors Think About It

Diagnostics, Год журнала: 2025, Номер 15(2), С. 168 - 168

Опубликована: Янв. 13, 2025

Background: Artificial intelligence (AI) has recently made unprecedented contributions in every walk of life, but it not been able to work its way into diagnostic medicine and standard clinical practice yet. Although data scientists, researchers, medical experts have working the direction designing developing computer aided diagnosis (CAD) tools serve as assistants doctors, their large-scale adoption integration healthcare system still seems far-fetched. Diagnostic radiology is no exception. Imagining techniques like magnetic resonance imaging (MRI), computed tomography (CT), positron emission (PET) scans widely very effectively employed by radiologists neurologists for differential diagnoses neurological disorders decades, yet AI-powered systems analyze such incorporated operating procedures systems. Why? It absolutely understandable that medicine, precious human lives are on line, hence there room even tiniest mistakes. Nevertheless, with advent explainable artificial (XAI), old-school black boxes deep learning (DL) unraveled. Would XAI be turning point finally embrace AI radiology? This review a humble endeavor find answers these questions. Methods: In this review, we present journey recognize, preprocess, brain MRI various disorders, special emphasis CAD embedded explainability. A comprehensive literature from 2017 2024 was conducted using host databases. We also domain experts’ opinions summarize challenges up ahead need addressed order fully exploit tremendous potential application diagnostics humanity. Results: Forty-seven studies were summarized tabulated information about technology datasets employed, along performance accuracies. The strengths weaknesses discussed. addition, seven around world presented guide engineers scientists tools. Conclusions: Current research observed focused enhancement accuracies DL regimens, less attention being paid authenticity usefulness explanations. shortage ground truth explainability observed. Visual explanation methods found dominate; however, they might enough, more thorough professor-like explanations would required build trust professionals. Special factors legal, ethical, safety, security issues can bridge current gap between routine practice.

Язык: Английский

---

Explaining explainability: The role of XAI in medical imaging

European Journal of Radiology, Год журнала: 2024, Номер 173, С. 111389 - 111389

Опубликована: Фев. 23, 2024

Язык: Английский

---

The Evolution of Artificial Intelligence in Medical Imaging: From Computer Science to Machine and Deep Learning

Cancers, Год журнала: 2024, Номер 16(21), С. 3702 - 3702

Опубликована: Ноя. 1, 2024

Artificial intelligence (AI), the wide spectrum of technologies aiming to give machines or computers ability perform human-like cognitive functions, began in 1940s with first abstract models intelligent machines. Soon after, 1950s and 1960s, machine learning algorithms such as neural networks decision trees ignited significant enthusiasm. More recent advancements include refinement algorithms, development convolutional efficiently analyze images, methods synthesize new images. This renewed enthusiasm was also due increase computational power graphical processing units availability large digital databases be mined by networks. AI soon applied medicine, through expert systems designed support clinician's later for detection, classification, segmentation malignant lesions medical A prospective clinical trial demonstrated non-inferiority alone compared a double reading two radiologists on screening mammography. Natural language processing, recurrent networks, transformers, generative have both improved capabilities making an automated images moved domains, including text analysis electronic health records, image self-labeling, self-reporting. The open-source free libraries, well powerful computing resources, has greatly facilitated adoption deep researchers clinicians. Key concerns surrounding healthcare need trials demonstrate efficacy, perception tools 'black boxes' that require greater interpretability explainability, ethical issues related ensuring fairness trustworthiness systems. Thanks its versatility impressive results, is one most promising resources frontier research applications particular oncological applications.

Язык: Английский

---

Expanded AI learning: AI as a Tool for Human Learning

Academic Radiology, Год журнала: 2025, Номер unknown

Опубликована: Апрель 1, 2025

Язык: Английский

---

Neuroimage analysis using artificial intelligence approaches: a systematic review

Medical & Biological Engineering & Computing, Год журнала: 2024, Номер 62(9), С. 2599 - 2627

Опубликована: Апрель 26, 2024

Язык: Английский

---

A systematic review on deep learning based methods for cervical cell image analysis

Neurocomputing, Год журнала: 2024, Номер unknown, С. 128630 - 128630

Опубликована: Сен. 1, 2024

Язык: Английский

---

Responsibility Gap(s) Due to the Introduction of AI in Healthcare: An Ubuntu-Inspired Approach

Science and Engineering Ethics, Год журнала: 2024, Номер 30(4)

Опубликована: Авг. 1, 2024

Due to its enormous potential, artificial intelligence (AI) can transform healthcare on a seemingly infinite scale. However, as we continue explore the immense potential of AI, it is vital consider ethical concerns associated with development and deployment. One specific concern that has been flagged in literature responsibility gap (RG) due introduction AI healthcare. When use an algorithm or system results negative outcome for patient(s), whom should be assigned? Although concept RG was introduced Anglo-American European philosophy, this paper aims broaden debate by providing Ubuntu-inspired perspective RG. Ubuntu, deeply rooted African calls collective responsibility, offers uniquely forward-looking approach address alleged caused An serve valuable guide tool when addressing Incorporating Ubuntu into ethics discourse contribute more responsible integration

Язык: Английский

---

Differences in technical and clinical perspectives on AI validation in cancer imaging: mind the gap!

European Radiology Experimental, Год журнала: 2025, Номер 9(1)

Опубликована: Янв. 15, 2025

Abstract Good practices in artificial intelligence (AI) model validation are key for achieving trustworthy AI. Within the cancer imaging domain, attracting attention of clinical and technical AI enthusiasts, this work discusses current gaps strategies, examining existing that common or variable across groups (TGs) (CGs). The is based on a set structured questions encompassing several topics, addressed to professionals working medical imaging. A total 49 responses were obtained analysed identify trends patterns. While TGs valued transparency traceability most, CGs pointed out importance explainability. Among topics where may benefit from further exposure stability robustness checks, mitigation fairness issues. On other hand, seemed more reluctant towards synthetic data would cross-validation techniques, segmentation metrics. Topics emerging open utility, capability, adoption trustworthiness. These findings strategies guide creation guidelines necessary training next generation with healthcare contribute bridging any technical-clinical gap validation. Relevance statement This study recognised understanding applying helped promote trust interdisciplinary teams researchers. Key Points Clinical researchers emphasise interpretability, external diverse data, bias awareness In research, prioritise explainability, while focus traceability, see potential datasets. Researchers advocate greater homogenisation Graphical

Язык: Английский

---

Multimodal deep learning fusion of ultrafast-DCE MRI and clinical information for breast lesion classification

Computers in Biology and Medicine, Год журнала: 2025, Номер 188, С. 109721 - 109721

Опубликована: Фев. 19, 2025

Breast cancer is the most common worldwide, and magnetic resonance imaging (MRI) constitutes a very sensitive technique for invasive detection. When reviewing breast MRI examination, clinical radiologists rely on multimodal information, composed of data but also information not present in images such as information. Most machine learning (ML) approaches are well suited data. However, attention-based architectures, Transformers, flexible therefore good candidates integrating The aim this study was to develop evaluate novel deep (DL) model combining ultrafast dynamic contrast-enhanced (UF-DCE) images, lesion characteristics classification. From 2019 2023, UF-DCE radiology reports 240 patients were retrospectively collected from single center annotated. Imaging constituted volumes interest (VOI) extracted around segmented lesions. Non-imaging both (categorical) geometrical (scalar) Clinical annotated associated their corresponding We compared diagnostic performances traditional ML methods non-imaging data, an image based DL architecture, Transformer-based Multimodal Sieve Transformer with Vision encoder (MMST-V). final dataset included 987 lesions (280 benign, 121 malignant lesions, 586 benign lymph nodes) 1081 reports. For classification scalar had greater influence (Area under receiver operating characteristic curve (AUROC) = 0.875 ± 0.042) than categorical (AUROC 0.680 0.060). MMST-V achieved better 0.928 0.027) 0.900 0.045), only 0.863 0.025). proposed adaptative approach that can consider redundant provided by It demonstrated unimodal methods. Results highlight combination patient detailed additional knowledge enhances MRI.

Язык: Английский

---