Large language models (LLMs), represented by ChatGPT, have garnered significant attention due to their powerful capabilities in understanding and generating human language. Research on the application of LLMs across various medical tasks has shown a vigorous development trend. This review aims to outline the development and clinical applications of LLMs, with a focus on the primary tasks of medical intelligent assistants, including their associated opportunities and challenges. At the technical level, we provide a detailed explanation of the architecture and training processes of existing medical LLMs, and summarize the general technical steps for adapting large models to the healthcare domain. At the application level, we introduce the main tasks of medical intelligent assistants from both healthcare provider- and patient-oriented perspectives, andcompare the performance of different LLMs across various medical tasks to illustrate their unique advantages and limitations in medical applications.

Purpose To explore the application value of combining 18F-FDG PET images with interpretable deep learning radiomics(IDLR)models in the differential diagnosis of primary Parkinson's disease(IPD)and atypical Parkinson's syndrome.Materials and Methods This cross-sectional study was conducted using the Parkinson's Disease PET Imaging Benchmark Database from Huashan Hospital,Fudan University from March 2015 to February 2023.A total of 330 Parkinson's disease patients underwent 18F-FDG PET imaging,both 18F-FDG PET imaging and clinical scale information were collected for all subjects.The study included two cohorts,a training group(n=270)and a testing group(n=60),with a total of 211 cases in the IPD group,59 cases in the progressive supranuclear palsy(PSP)group,and a group of 60 patients with multiple system atrophy(MSA).The clinical information between different groups were compared.An IDLR model was developed to extract feature indicators.Under the supervision of radiomics features,IDLR features were selected from the features collected by neural network extractors,and a binary support vector machine model was constructed for the selected features in images of in testing group.The constructed IDLR model,traditional radiomics model and standard uptake ratio model were separately used to calculate the performance metrics and area under curve values of deep learning models for pairwise classification between IPD/PSP/MSA groups.The study conducted independent classification and testing in two cohorts using 100 10-fold cross-validation tests.Brain-related regions of interest were displayed through feature mapping,using gradient weighted class activation maps to highlight and visualize the most relevant information in the brain.The output heatmaps of different disease groups were examined and compared with clinical diagnostic locations.Results The IDLR model showed promising results for differentiating between Parkinson's syndrome patients.It achieved the best classification performance and had the highest area under the curve values compared to other comparative models such as the standard uptake ratio model(Z=1.22-3.23,all P<0.05),and radiomics model(Z=1.31-2.96,all P<0.05).The area under the curve values for the IDLR model in differentiating MSA and IPD were 0.935 7,for MSA and PSP were 0.975 4,for IPD and PSP were 0.982 5 in the test set.The IDLR model also showed consistency between its filtered feature maps and the visualization of gradient-weighted class activation mapping slice thermal maps in the radiomics regions of interest.Conclusion The IDLR model has the potential for differential diagnosis between IPD and atypical Parkinson's syndrome in 18F-FDG PET images.

Objective:To compare the effects of different reference brain regions on the semi-quantitative SUV ratio (SUVR) of 18F-Florzolotau PET images of Alzheimer′s disease (AD). Methods:The 18F-Florzolotau PET images of 28 (13 males, 15 females, age (57.3±9.5) years) normal controls (NC), 19 patients (4 males, 15 females, age (73.3±7.3) years) with β-amyloid (Aβ)-positive mild cognitive impairment (MCI) and 40 patients (19 males, 21 females, age (61.9±9.1) years) with AD were collected from Huashan Hospital, Fudan University between November 2018 and July 2020. Six semi-quantitative reference brain regions were defined, including whole cerebellum (WC), cerebellar gray matter (GM), cerebellar white matter (WM), parametric estimation of reference signal intensity (PERSI), WC after partial volume correction (WC_pvc), cerebellar GM after partial volume correction (GM_pvc). SUVR was calculated for 14 ROIs, which included the whole brain defined by the automated anatomical labeling (AAL) template, fusiform, inferior temporal, lingual, middle temporal, occipital, parahippocampal, parietal, posterior cingulate, precuneus defined by the AAL template, and Meta ROI composed of the above brain regions, and braak_Ⅰ-Ⅱ, braak_Ⅲ-Ⅳ, braak_Ⅴ-Ⅵ defined by the Desikan Killiany template. AUC was used to evaluate the classification ability of SUVR, and the correlation between SUVR and clinical scale scores were assessed by Spearman rank correlation analysis. Results:The SUVRs of most brain regions showed a steady upward trend in the AD disease spectrum. In the classification task of NC and MCI, the overall performance of SUVR based on WC_pvc was relatively optimal (AUCs: 0.975-1.000). In the classification task of NC and AD, SUVRs of 10 ROIs based on the WC_pvc method showed the relatively best performance (AUCs: 0.976-1.000). The correlation between SUVR of fusiform based on cerebellar WM and mini-mental state examination (MMSE) score was the strongest ( rs=-0.72, P<0.001), and the SUVR of precuneus based on WC_pvc showed the strongest correlation with clinical dementia rating (CDR) score ( rs=0.78, P<0.001). Conclusion:The SUVR based on WC_pvc method performs well in classification and correlation tasks, and is recommended to be used in semi-quantification of 18F-Florzolotau PET images of AD.

Objective:To perform harmonization based on the ComBat method for PET brain imaging scanned by different types of scanners from the same manufacturer and explored its effect on center effect.Methods:The three-dimensional (3D) Hoffman brain model was scanned by two different PET/CT instruments (Siemens Biograph64 TruePoint and Biograph128 mCT). Fourteen healthy subjects (8 males, 6 females, age: (57.7±9.5) years) underwent 18F-FDG PET/CT on Siemens Biograph64 TruePoint and 12 healthy subjects (9 males, 3 females, age: (55.8±10.5) years) underwent 18F-FDG PET/CT on Siemens Biograph128 mCT (all from Huashan Hospital, Fudan University; from November 2020 to March 2023). The whole brain was divided into 116 brain regions based on the anatomical automatic labeling (AAL) brain template. The ComBat method was applied to harmonized the PET data from brain model and healthy subjects. Mann-Whitney U test was performed on the radioactive counts and SUV ratios (SUVR) before and after homogenization acquired by both PET/CT instruments. Voxel-based statistical parametric mapping (SPM) independent-sample t test was also performed on data of healthy subjects. Results:In 3D Hoffman brain model, radioactivity counts (5 590.33(4 961.67, 6 102.95) vs 6 116.03(5 420.97, 6 660.66); z=-9.35, P<0.001) and SUVR (1.35(1.19, 1.47) vs 1.37(1.21, 1.49); z=-3.63, P<0.001) were significantly different between the two PET/CT scanners before harmonization and not after harmonization (radioactivity counts: 5 845.95(5 192.68, 6 378.63) vs 5 859.17(5 193.84, 6 380.52); SUVR: 1.35(1.20, 1.48) vs 1.36(1.20, 1.49); both z=-0.68, both P=0.498). In the healthy subjects, radioactive counts in 19 brain regions (12 422.78(11 181.60, 13 424.28)-18 166.40(15 882.80, 18 666.27); z values: from -3.24 to -2.06, all P<0.05) and SUVR in 40 brain regions (1.46(1.41, 1.52)-2.28(2.16, 2.36); z values: from -3.65 to -1.70, all P<0.05) were significantly different between the two scanners before harmonization, while after homogenization there were no statistical differences for all 116 brain regions (radioactivity counts: 9 243.55(8 502.38, 9 854.87)-20 419.60(19 931.51, 21 179.43); z values: from -0.72 to 0, all P>0.05; SUVR: 1.04(1.01, 1.09)-2.32(2.24, 2.40); z values: from -0.82 to 0, all P>0.05). SPM showed that significant differences of glucose metabolism in the cerebral cortex, basal ganglia, midbrain and cerebellum were found in healthy subjects between the two PET/CT scanners before homogenization, and brain regions with obvious differences reduced after homogenization. Conclusion:ComBat harmonization method is efficient at removing the center effect among different types of PET/CT scanners from the same manufacturer and may provide a simple and easy-to-implement homogenization for multicenter brain imaging studies.

Objective:A voxel-level quantification method based on the tau IQ algorithm and Braak staging, excluding β-amyloid (Aβ) imaging, was developed to achieve specific tau quantification. Methods:This cross-sectional study included 92 subjects (35 males, 57 females; age (62.9±10.4) years) from the Nuclear Medicine/PET Center of Huashan Hospital, Fudan University between November 2018 and July 2020. The cohort comprised 28 cognitively normal (CN) individuals, 20 patients with mild cognitive impairment (MCI), and 44 patients with Alzheimer′s disease (AD). All participants underwent 18F-florzolotau PET imaging, Mini-Mental State Examination (MMSE), and Clinical Dementia Rating (CDR) scoring. A longitudinal tau dataset was constructed based on Braak staging. Voxel-level logistic regression fitting provided a baseline matrix, decomposed via least squares to yield the Tau load coefficient. One-way analysis of variance (with post hoc Tukey) was used to compare Tau load and SUV ratio (SUVR) among groups. ROC curve analysis was used to evaluate classification between CN, MCI and AD. Spearman rank correlation was used to assess the relationships between Tau load, SUVR, and MMSE scores or CDR scores. Results:The Tau load in the CN group was close to 0 and significantly lower than that in the MCI and AD groups ( F=55.03, P<0.001; post hoc tests all P<0.001). Significant differences were also observed in the SUVR across all ROIs ( F values: 36.46-55.38, all P<0.001). Compared to SUVR, Tau load demonstrated greater intergroup differences. In ROC curve analyses between each pair of CN, MCI, and AD groups, Tau load consistently achieved the highest AUC (0.754-1.000). Both Tau load and SUVR for each ROI were negatively correlated with MMSE scores ( rs values: from -0.698 to -0.583, all P<0.05) and positively correlated with CDR scores ( rs values: 0.648-0.783, all P<0.05), with Tau load showing the highest absolute correlation coefficients. Conclusion:Compared to the traditional semi-quantitative SUVR method, the Braak-tau IQ algorithm does not require a specific reference brain region to achieve specific tau quantification.

We reported a case of a patient dignosed as tuberculosis and severe malnutrition with coronavirus disease 2019 （critical type） treated with a combination of Chinese medicine and Western medicine. Through the retrospective analysis of the diagnosis and treatment process of this patient， on the basis of Chinese medicine's understanding of the etiology and pathogenesis of “old state” and “deficient state”， the critical coronavirus pneumonia combined with pulmonary tuberculosis and severe malnutrition was mostly due to the physical condition and the invasion of epidemic toxin， resulting in dysfunctions of the internal organs such as the lungs， spleens， kidneys and other organs. Based on the understanding of the cause and mechanism of the coronavirus disease， the treatment combined Chinese and Western medical therapies was given. The western medicine was used with the main treatments of oxygen therapy， anti-viral， intestinal nutritional support， and anti-coagulation， while the Chinese medicine was used by tonifying qi， blood， yin， and yang， warming yang and dissipating cold， and clearing heat and dampness， then tonifying qi， nourishing yin and eliminating heat， in which tonifying middle and replenishing qi ran through the whole process. The integrated treatment eventually improved the patient's symptoms and accelerated the negative conversion of nucleic acid of the coronavirus.

Presynaptic dopaminergic PET imaging is a useful method for the diagnosis of parkinsonism. Based on the expert consensus on operation and clinical application of dopamine transporter brain PET imaging technology published in 2020, this paper further recommends the relevant elements of result interpretation of presynaptic dopaminergic PET imaging.

Hepatitis B virus (HBV) infection is a global health problem. Animal models are important for the study of the HBV infection mechanism. In the study related to the mouse model of HBV infection, the researchers have established a variety of mouse models, including transgenic, plasmid hydrodynamic injection, virus vector transfection, cccDNA cycle simulation, human and mouse liver chimerism, and liver/immune dual humanization, according to the characteristics of HBV infection. Herein, the research progress of these models is summarized. Notably, the application of these models can further clarify the mechanism of HBV infection under the conditions of a specific immune response in vivo and lay the foundation for the development of new antiviral drugs and immunotherapy for HBV infection.

Objective:To investigate whether the presynaptic dopamine neuronal depletion in different striatal subregions predicts future development of wearing-off (WO) in Parkinson′s disease (PD) patients.Methods:A retrospective longitudinal study included 57 PD patients who were referred to the Department of Neurology of Xinhua Hospital Affiliated to Shanghai Jiao Tong University School of Medicine from January 2019 to September 2020, and completed 11C-2β-carbomethoxy-3β-(4-fluorophenyl) tropane dopamine transporter (DAT) positron emission tomography scans at the initial evaluation and received dopaminergic drugs for at least 12 months during follow-up. The time of starting dopaminergic drug treatment and the occurrence of WO were recorded. After adjusting for clinical related factors, the predictive value of DAT uptake and related parameters in striatal subregions for WO was evaluated by Cox proportional hazards model. Results:During a median follow-up period of 23 months, 10 patients (18.18%) developed WO. Patients with WO exhibited less DAT uptake in the caudate nucleus and anterior putamen nucleus (0.66±0.52 vs 1.08±0.42, t=2.76, P=0.008 and 0.66±0.20 vs 0.87±0.28, t=2.27, P=0.027 respectively), especially in these subregions contralateral to the less-affected side of the body, compared to those without WO. Cox proportional hazard models revealed that after adjusting for gender, age, course of disease, baseline Unified Parkinson′s Disease Rating Scale Ⅲ score and increment of levodopa equivalent dosage, the lower the DAT uptake of the caudate ipsilateral to the less-affected side of the body ( HR=0.20, 95% CI 0.07-0.63, P=0.006), as well as the lower the DAT uptake of the caudate nucleus and posterior putamen nucleus ( HR=0.28, 95% CI 0.11-0.69, P=0.006 and HR=0.08, 95% CI 0.01-0.64, P=0.018 respectively) and the higher the ratio of putamen/caudate contralateral to the less-affected side of the body ( HR=2.33, 95% CI 1.02-5.33, P=0.045), the higher the risk of WO. Conclusion:The presynaptic dopamine neuronal loss, particularly bilateral caudate nucleus dopaminergic depletion at the early stage, has predictive value of development of WO in PD.

Objective:To explore the potential application of combining 18F-FDG PET imaging and radiomics in the diagnosis of Parkinson′s disease (PD) and atypical parkinsonian syndromes (APS). Methods:A total of 154 subjects of two cohorts (training set and validation set) were enrolled from Huashan Hospital, Fudan University from March 2015 to August 2020 in this cross-sectional study, including 40 normal controls (NC; 23 males and 17 females, age: (60.2±10.5) years), 40 PD patients (20 males and 20 females, age: (64.7±6.3) years), 40 progressive supranuclear palsy (PSP) patients (20 males and 20 females, age: (64.1±5.9) years), and 34 multiple system atrophy (MSA) patients (19 males and 15 females, age: (65.0±9.2) years). 18F-FDG PET images and clinical scale were selected, and one-way analysis of variance was used to compare differences of clinical scale among groups. Radiomic features extraction and feature selection were carried out. Two and three classification models were constructed based on logistic regression, and the ROC curves of clinical model, radiomics model and combined model were calculated. Independent classification tests were conducted 100 times with 5-fold cross validation in two cohorts. Results:There were significant differences in the scores of unified PD Rating Scale (UPDRS) and Hoehn-Yahr rating scale (H&Y) among different groups in cohort 1 and cohort 2 respectively ( F values: 4.83-17.95, all P<0.05). A total of 2 444 imaging features were extracted from each subject, and after features selection, 15 features for classification were obtained. In the two classification experiment, the AUCs of the three models in binary classification of PD/MSA/PSP/NC group were 0.56-0.68, 0.74-0.93 and 0.72-0.93, respectively. The classification effects of the radiomics model were significantly better than those of the clinical model ( z values: 1.71-2.85, all P<0.05). In the three classification experiment, the sensitivity of the radiomics model reached 80%, 80% and 77% for PD, MSA and PSP, respectively. Conclusion:18F-FDG imaging combined with radiomics has potential in the diagnosis of PD and APS.