Artificial Intelligence Model for a Distinction between Early-Stage Gastric Cancer Invasive Depth T1a and T1b.

Background: Endoscopic submucosal dissection (ESD) is a widely accepted treatment for patients with mucosa (T1a) disease without lymph node metastasis. However, the inconsistency of inspection quality of tumor staging under the standard tool combining endoscopic ultrasound (EUS) with computed tomography (CT) scanning makes it restrictive. Methods: We conducted a study using data augmentation and artificial intelligence (AI) to address the early gastric cancer (EGC) staging problem. The proposed AI model simplifies early cancer treatment by eliminating the need for ultrasound or other staging methods. We developed an AI model utilizing data augmentation and the You-Only-Look-Once (YOLO) approach. We collected a white-light image dataset of 351 stage T1a and 542 T1b images to build, test, and validate the model. An external white-light images dataset that consists of 47 T1a and 9 T1b images was then collected to validate our AI model. The result of the external dataset validation indicated that our model also applies to other peer health institutes. Results: The results of k-fold cross-validation using the original dataset demonstrated that the proposed model had a sensitivity of 85.08% and an average specificity of 87.17%. Additionally, the k-fold cross-validation model had an average accuracy rate of 86.18%; the external data set demonstrated similar validation results with a sensitivity of 82.98%, a specificity of 77.78%, and an overall accuracy of 82.14%. Conclusions: Our findings suggest that the AI model can effectively replace EUS and CT in early GC staging, with an average validation accuracy rate of 86.18% for the original dataset from Linkou Cheng Gun Memorial Hospital and 82.14% for the external validation dataset from Kaohsiung Cheng Gun Memorial Hospital. Moreover, our AI model's accuracy rate outperformed the average EUS and CT rates in previous literature (around 70%).

A colonial serrated polyp classification model using white-light ordinary endoscopy images with an artificial intelligence model and TensorFlow chart.

In this study, we implemented a combination of data augmentation and artificial intelligence (AI) model-Convolutional Neural Network (CNN)-to help physicians classify colonic polyps into traditional adenoma (TA), sessile serrated adenoma (SSA), and hyperplastic polyp (HP). We collected ordinary endoscopy images under both white and NBI lights. Under white light, we collected 257 images of HP, 423 images of SSA, and 60 images of TA. Under NBI light, were collected 238 images of HP, 284 images of SSA, and 71 images of TA. We implemented the CNN-based artificial intelligence model, Inception V4, to build a classification model for the types of colon polyps. Our final AI classification model with data augmentation process is constructed only with white light images. Our classification prediction accuracy of colon polyp type is 94%, and the discriminability of the model (area under the curve) was 98%. Thus, we can conclude that our model can help physicians distinguish between TA, SSA, and HPs and correctly identify precancerous lesions such as TA and SSA.

Development of a Machine Learning Model to Predict the Use of Surgery in Patients With Rheumatoid Arthritis.

OBJECTIVE: One in five patients with rheumatoid arthritis (RA) rely on surgery to restore joint function. However, variable response to disease-modifying antirheumatic drugs (DMARDs) complicates surgical planning, and it is difficult to predict which patients may ultimately require surgery. We used machine learning to develop predictive models for the likelihood of undergoing an operation related to RA and which type of operation patients who require surgery undergo. METHODS: We used electronic health record data to train two extreme gradient boosting machine learning models. The first model predicted patients' probabilities of undergoing surgery ≥5 years after their initial clinic visit. The second model predicted whether patients who underwent surgery would undergo a major joint replacement versus a less intensive procedure. Predictors included demographics, comorbidities, and medication data. The primary outcome was model discrimination, measured by area under the receiver operating characteristic curve (AUC). RESULTS: We identified 5,481 patients, of whom 278 (5.1%) underwent surgery. There was no significant difference in the frequency of DMARD or steroid prescriptions between patients who did and did not have surgery, though nonsteroidal anti-inflammatory drug prescriptions were more common among patients who did have surgery (P = 0.03). The model predicting use of surgery had an AUC of 0.90 ± 0.02. The model predicting type of surgery had an AUC of 0.58 ± 0.10. CONCLUSIONS: Predictive models using clinical data have the potential to facilitate identification of patients who may undergo rheumatoid-related surgery, but not what type of procedure they will need. Integrating similar models into practice has the potential to improve surgical planning.

Cancer survivorship and risk of pregnancy complications, adverse obstetric outcomes, and maternal morbidities in female adolescents and young adults: a nationwide population-based study from Taiwan.

BACKGROUND: Cancer treatment in female adolescent and young adult (AYA) cancer survivors (i.e., those diagnosed between 15 and 39 years of age) may adversely affect multiple bodily functions, including the reproductive system. METHODS: We initially assembled a retrospective, nationwide population-based cohort study by linking data from two nationwide Taiwanese data sets. We subsequently identified first pregnancies and singleton births to AYA cancer survivors (2004-2018) and select AYA without a previous cancer diagnosis matched to AYA cancer survivors for maternal age and infant birth year. RESULTS: The study cohort consisted of 5151 and 51,503 births to AYA cancer survivors and matched AYA without a previous cancer diagnosis, respectively. The odds for overall pregnancy complications (odds ratio [OR], 1.09; 95% confidence interval [CI], 1.01-1.18) and overall adverse obstetric outcomes (OR, 1.07; 95% CI, 1.01-1.13) were significantly increased in survivors compared with matched AYA without a previous cancer diagnosis. Specifically, cancer survivorship was associated with an increased risk of preterm labour, labour induction, and threatened abortion or threatened labour requiring hospitalisation. CONCLUSIONS: AYA cancer survivors are at increased risk for pregnancy complications and adverse obstetric outcomes. Efforts to integrate individualised care into clinical guidelines for preconception and prenatal care should be thoroughly explored.

Automatic and human level Graf's type identification for detecting developmental dysplasia of the hip.

BACKGROUND: Developmental dysplasia of the hip (DDH) is a common congenital disorder that may lead to hip dislocation and requires surgical intervention if left untreated. Ultrasonography is the preferred method for DDH screening; however, the lack of experienced operators impedes its application in universal neonatal screening. METHODS: We developed a deep neural network tool to automatically register the five keypoints that mark important anatomical structures of the hip and provide a reference for measuring alpha and beta angles following Graf's guidelines, which is an ultrasound classification system for DDH in infants. Two-dimensional (2D) ultrasonography images were obtained from 986 neonates aged 0-6 months. A total of 2406 images from 921 patients were labeled with ground truth keypoints by senior orthopedists. RESULTS: Our model demonstrated precise keypoint localization. The mean absolute error was approximately 1 mm, and the derived alpha angle measurement had a correlation coefficient of R = 0.89 between the model and ground truth. The model achieved an area under the receiver operating characteristic curve of 0.937 and 0.974 for classifying alpha <60° (abnormal hip) and <50° (dysplastic hip), respectively. On average, the experts agreed with 96% of the inferenced images, and the model could generalize its prediction on newly collected images with a correlation coefficient higher than 0.85. CONCLUSIONS: Precise localization and highly correlated performance metrics suggest that the model can be an efficient tool for assisting DDH diagnosis in clinical settings.

Can a Deep Learning Algorithm Improve Detection of Occult Scaphoid Fractures in Plain Radiographs? A Clinical Validation Study.

BACKGROUND: Occult scaphoid fractures on initial radiographs of an injury are a diagnostic challenge to physicians. Although artificial intelligence models based on the principles of deep convolutional neural networks (CNN) offer a potential method of detection, it is unknown how such models perform in the clinical setting. QUESTIONS/PURPOSES: (1) Does CNN-assisted image interpretation improve interobserver agreement for scaphoid fractures? (2) What is the sensitivity and specificity of image interpretation performed with and without CNN assistance (as stratified by type: normal scaphoid, occult fracture, and apparent fracture)? (3) Does CNN assistance improve time to diagnosis and physician confidence level? METHODS: This survey-based experiment presented 15 scaphoid radiographs (five normal, five apparent fractures, and five occult fractures) with and without CNN assistance to physicians in a variety of practice settings across the United States and Taiwan. Occult fractures were identified by follow-up CT scans or MRI. Participants met the following criteria: Postgraduate Year 3 or above resident physician in plastic surgery, orthopaedic surgery, or emergency medicine; hand fellows; and attending physicians. Among the 176 invited participants, 120 completed the survey and met the inclusion criteria. Of the participants, 31% (37 of 120) were fellowship-trained hand surgeons, 43% (52 of 120) were plastic surgeons, and 69% (83 of 120) were attending physicians. Most participants (73% [88 of 120]) worked in academic centers, whereas the remainder worked in large, urban private practice hospitals. Recruitment occurred between February 2022 and March 2022. Radiographs with CNN assistance were accompanied by predictions of fracture presence and gradient-weighted class activation mapping of the predicted fracture site. Sensitivity and specificity of the CNN-assisted physician diagnoses were calculated to assess diagnostic performance. We calculated interobserver agreement with the Gwet agreement coefficient (AC1). Physician diagnostic confidence was estimated using a self-assessment Likert scale, and the time to arrive at a diagnosis for each case was measured. RESULTS: Interobserver agreement among physicians for occult scaphoid radiographs was higher with CNN assistance than without (AC1 0.42 [95% CI 0.17 to 0.68] versus 0.06 [95% CI 0.00 to 0.17], respectively). No clinically relevant differences were observed in time to arrive at a diagnosis (18 ± 12 seconds versus 30 ± 27 seconds, mean difference 12 seconds [95% CI 6 to 17]; p < 0.001) or diagnostic confidence levels (7.2 ± 1.7 seconds versus 6.2 ± 1.6 seconds; mean difference 1 second [95% CI 0.5 to 1.3]; p < 0.001) for occult fractures. CONCLUSION: CNN assistance improves physician diagnostic sensitivity and specificity as well as interobserver agreement for the diagnosis of occult scaphoid fractures. The differences observed in diagnostic speed and confidence is likely not clinically relevant. Despite these improvements in clinical diagnoses of scaphoid fractures with the CNN, it is unknown whether development and implementation of such models is cost effective. LEVEL OF EVIDENCE: Level II, diagnostic study.

Development and Validation of a Novel Score for Predicting Long-Term Mortality after an Acute Ischemic Stroke.

BACKGROUND: Long-term mortality prediction can guide feasible discharge care plans and coordinate appropriate rehabilitation services. We aimed to develop and validate a prediction model to identify patients at risk of mortality after acute ischemic stroke (AIS). METHODS: The primary outcome was all-cause mortality, and the secondary outcome was cardiovascular death. This study included 21,463 patients with AIS. Three risk prediction models were developed and evaluated: a penalized Cox model, a random survival forest model, and a DeepSurv model. A simplified risk scoring system, called the C-HAND (history of Cancer before admission, Heart rate, Age, eNIHSS, and Dyslipidemia) score, was created based on regression coefficients in the multivariate Cox model for both study outcomes. RESULTS: All experimental models achieved a concordance index of 0.8, with no significant difference in predicting poststroke long-term mortality. The C-HAND score exhibited reasonable discriminative ability for both study outcomes, with concordance indices of 0.775 and 0.798. CONCLUSIONS: Reliable prediction models for long-term poststroke mortality were developed using information routinely available to clinicians during hospitalization.

Effect of febuxostat on adverse events and mortality in gout in Taiwan: An interrupted time series analysis.

OBJECTIVES: To evaluate the influence of febuxostat on adverse events and mortality in gout. METHODS: We retrospectively enrolled patients with newly diagnosed gout and prescribed urate-lowering therapy between 2006 and 2017 from the Taiwan National Health Insurance Database. These patients were divided into 2 groups: with and without comorbidities (n = 294 847 and 194 539). An interrupted time series analysis with adjustments for demographics, comorbidities, and comedication by propensity score-based stabilized weights was used to compare the trend of adverse events and mortality before vs after febuxostat was introduced in 2012. RESULTS: The proportion of febuxostat use gradually increased from 0% in 2012 to 30% in those with comorbidities and 10% in those without comorbidities in 2017. Allopurinol use decreased from 30% in 2012 to 10% in 2017. The slope of the 1-year incidence rate of Stevens-Johnson syndrome (SJS) or toxic epidermal necrolysis (TEN) (per 10 000 patients) significantly reduced after 2012 in those with and without comorbidities (-0.375 per quarter, P = .015 and -.253 per quarter, P = .049). The slope of the 3-year incidence rate of acute myocardial infarction (AMI) (per 1000 patients), percutaneous coronary intervention (PCI) (per 1000 patients), and all-cause mortality (per 100 patients) significantly increased after 2012 in those with comorbidities (+0.207 per quarter, P = .013; +.389 per quarter, P = .002; +.103 per quarter, P = .001). CONCLUSIONS: Febuxostat may reduce SJS and TEN in all gout patients but increase AMI, PCI, and all-cause mortality in gout patients with comorbidities.

Reduced versus maximum tolerated methotrexate dose concomitant with adalimumab in patients with rheumatoid arthritis (MIRACLE): a randomised, open-label, non-inferiority trial.

BACKGROUND: Efficacy of combination therapy with methotrexate and biological disease-modifying antirheumatic drugs is well established in the management of patients with rheumatoid arthritis; however, the optimal dose of methotrexate to administer with a tumour necrosis factor inhibitor remains unclear. We aimed to clarify the efficacy and safety of adalimumab combined with reduced methotrexate dose compared with the maximum tolerated methotrexate dose in patients with rheumatoid arthritis and an inadequate response to methotrexate monotherapy. METHODS: In this open-label, randomised controlled trial, we recruited methotrexate-naive patients with rheumatoid arthritis and a disease duration of less than 2 years across 24 secondary or tertiary care hospitals across Japan, South Korea, and Taiwan. At initiation, methotrexate was given orally and increased to the maximum tolerated dose by week 12. Patients who did not achieve remission on the basis of the Simplified Disease Activity Index (SDAI) at week 24 were randomly assigned (1:1) to receive adalimumab (40 mg biweekly) combined with a continued maximum tolerated dose of methotrexate or adalimumab combined with a reduced dose of methotrexate. The primary endpoint was non-inferiority of adalimumab plus reduced-dose methotrexate to adalimumab plus maximal-dose methotrexate based on SDAI remission at week 48, assessed in the modified full-analysis set with a pre-specified non-inferiority margin of -15%, based on a two-sided 90% CI. Adverse events were assessed in the safety analysis set. This trial is registered with ClinicalTrials.gov, NCT03505008 and has been completed. FINDINGS: From April 18, 2018, to June 2, 2020, from 323 patients screened, 300 were enrolled, and 291 patients were included in the full analysis set. The mean age was 57·7 years (SD 15·2), 217 (75%) were female, 74 (25%) were male, and all patients were of Asian ethnicity. The mean SDAI at study enrolment was 26·5 (SD 12·4). 52 patients discontinued the study before week 24 or at week 24 before randomisation. At week 24, 105 (36%) of 291 patients achieved remission and continued methotrexate monotherapy through week 48. 134 (46%) did not achieve remission at week 24 and were randomly assigned to receive adalimumab plus the maximum tolerated dose of methotrexate (n=68) or adalimumab plus reduced-dose methotrexate (n=66). Remission at week 48 was achieved in 25 (38%) of 66 and 27 (44%) of 61 patients, respectively, with an adjusted risk difference of 6·4% (90% CI -7·0 to 19·8), which met the non-inferiority margin of -15%. Adverse events after week 24 tended to be more frequent in the maximum tolerated dose group than in the reduced-dose group (24 [35%] vs 13 [20%], p=0·054). Between week 24 and 48, there were 14 serious adverse events (6 in the methotrexate monotherapy group, 5 in the adalimumab plus maximal-dose methotrexate, and 3 in the adalimumab plus reduced-dose methotrexate group), and no deaths. INTERPRETATION: The MIRACLE study showed that the efficacy of adalimumab combined with reduced methotrexate dose was not inferior to that with the maximum tolerated methotrexate dose, with a tendency to a better safety profile. FUNDING: Eisai.

Epidemiology and Mortality of Ovarian Cancer in Taiwan: A Population-Based Study.

Ovarian cancer is the second most common cause of death from gynecologic cancer. The aim of this study was to estimate the incidence of ovarian cancer and the trend of mortality in different histological subtypes of ovarian cancer in Taiwan. Patient information regarding ovarian cancer was provided by the Taiwan National Health Insurance database. The histological subtypes of ovarian cancer were retrieved from the Taiwan Cancer Registry database, while the survival rates were extracted from the National Death Registry database. In this population-based cohort study, the annual prevalence, incidence, and overall mortality of ovarian cancer during 2002-2015 were determined. The trend in the incidence and the mortality rate of different histologic subtypes were estimated using joinpoint regression analysis. It was found that age-standardized incidence of ovarian cancer increased from 9.46 in 2002 to 11.92 per 100,000 person-years in 2015, with an average annual percentage change of 2.0 (95% CI = 1.5-2.5). The 1-, 3-, and 5-year mortality rates of overall ovarian cancer declined progressively during the study period, especially the group of Charlson comorbidity index ≤ 1. Ovarian serous carcinoma was the most common histological subtype in Taiwan, comprising 30.9% of ovarian cancer patients in 2002-2015. This study provides valuable information for use in developing healthcare policies for ovarian cancer.

Deep Learning-Based Nuclear Morphometry Reveals an Independent Prognostic Factor in Mantle Cell Lymphoma.

Blastoid/pleomorphic morphology is associated with short survival in mantle cell lymphoma (MCL), but its prognostic value is overridden by Ki-67 in multivariate analysis. Herein, a nuclear segmentation model was developed using deep learning, and nuclei of tumor cells in 103 MCL cases were automatically delineated. Eight nuclear morphometric attributes were extracted from each nucleus. The mean, variance, skewness, and kurtosis of each attribute were calculated for each case, resulting in 32 morphometric parameters. Compared with those in classic MCL, 17 morphometric parameters were significantly different in blastoid/pleomorphic MCL. Using univariate analysis, 16 morphometric parameters (including 14 significantly different between classic and blastoid/pleomorphic MCL) emerged as significant prognostic factors. Using multivariate analysis, Biologic MCL International Prognostic Index (bMIPI) risk group (P = 0.025), low skewness of nuclear irregularity (P = 0.020), and high mean of nuclear irregularity (P = 0.047) emerged as independent adverse prognostic factors. Additionally, a morphometric score calculated from the skewness and mean of nuclear irregularity (P = 0.0038) was an independent prognostic factor in addition to bMIPI risk group (P = 0.025), and a summed morphometric bMIPI score was useful for risk stratification of patients with MCL (P = 0.000001). These results demonstrate, for the first time, that a nuclear morphometric score is an independent prognostic factor in MCL. It is more robust than blastoid/pleomorphic morphology and can be objectively measured.

A chemiresistive biosensor for detection of cancer biomarker in biological fluids using CVD-grown bilayer graphene.

A chemiresistive biosensor is described for simple and selective detection of miRNA-21. We developed chemical vapor deposition (CVD) and low-damage plasma treatment (LDPT)-treated bilayer graphene composite of graphene oxide/graphene (GO/GR) for the determination of a reliable biomarker. We have successfully overcome the self-limiting growth mechanism by using CVD method to grow more than one layer of graphene on copper foil. In addition, LDPT can be used to form GO/GR structures for chemiresistive biosensor applications. Due to the direct formation of BLGR (bilayer graphene), the coupling between graphene layers is theoretically superior to that of stacked BLGR, which is also confirmed by the blue shift of the characteristic peak of graphene in Raman spectroscopy. The shift is about double compared with that of stacked BLGR. Based on the results, the limit of detection for the target miRNA-21 was calculated to be 5.20 fM and detection rage is calculated as 100 fM to 10 nM, which is obviously better performance. Compared with previous work, this chemiresistive biosensor has good selectivity, and stability towards detection of miRNA-21. The ability to detect miRNA-21 in different biological fluids was almost identical to that in pH 7.4 phosphate-buffered saline (PBS). Thus, the proposed bilayer GO/GR of modified chemiresistive biosensor may potentially be applied to detect cancer cells in clinical examinations.

Comparative safety of Janus kinase inhibitors and tumor necrosis factor inhibitors in patients undergoing treatment for rheumatoid arthritis.

OBJECTIVES: Since 2010, biological disease-modifying antirheumatic drugs (bDMARDs) have been the dominant mode of treatment for rheumatoid arthritis (RA). However, the safety of DMARDs, such as tumor necrosis factor inhibitors (TNFis) and Janus kinase inhibitors (JAKis), in treating patients with RA is a concern. We compared the safety outcomes of JAKis and TNFis in RA patients in clinical settings. METHODS: Patients diagnosed with RA between 2015 and 2017 were identified from the Taiwan National Health Insurance Research Database and followed till 2018. Propensity score stabilized weighting was used to balance the baseline characteristics of the JAKis and TNFis groups. The incidences of safety outcomes, namely cardiovascular (CV) events, tuberculosis (TB), total hip replacement (THR), total knee replacement (TKR), and all-cause mortality, were compared between the 2 study groups. RESULTS: A total of 3179 patients with RA who were administered JAKis (n = 822) and TNFis (n = 2357) were included in this study. The mean follow-up duration was 2.02 years in the JAKis group and 2.10 in the TNFis group. All-cause mortality had the highest incidence rate, followed by TKR, THR, CV events, and TB. A lower incidence rate of the study outcomes was observed in the JAKis group than in the TNFis group but without statistical significance. CONCLUSION: Comparable safety issues and mortality rates were observed for JAKis and TNFis in RA patients treated in real-world settings.

Breast cancer-related lymphedema correlated with incidence of cellulitis and mortality.

BACKGROUND: This study investigated breast cancer-related lymphedema (BCRL) and its correlation with the incidence of cellulitis and mortality in the National Health Insurance (NHI) database in Taiwan. METHODS: Between 2004 and 2014, the NHI database of patients with breast cancer who underwent surgical procedures, adjuvant therapies, BCRL, cellulitis, and mortality were retrospectively reviewed. Cox regression was used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for incidence of BCRL and cellulitis in different treatment groups. The associations of BCRL with the incidence of cellulitis and mortality were further analyzed using the Kaplan-Meier curve. RESULTS: Among 100 301 patients, 5464 (5.4%) developed BCRL with a median onset of 1.3 years. At a mean follow-up of 4.77 years, the incidence of cellulitis in the BCRL group (12.7%, 694/5464 patients) was significantly higher than in the no-BCRL group (2.73%, 2589/94 837 patients) (HR: 3.74; 95% CI: 3.43-4.08; p < 0.0001). At a mean follow-up of 5.77 years, the mortality rate in the cellulitis group (34.21%, 1123/3283 patients) was significantly greater than in the no-cellulitis group (16.29%, 15 804/97 018 patients) (HR: 1.17; 95% CI: 1.1-1.24; p < 0.0001). CONCLUSIONS: BCRL had a significantly higher incidence of cellulitis and mortality.

Trends of adverse events and mortality after DMARDs in patients with rheumatoid arthritis: Interrupted time-series analysis.

OBJECTIVES: Patients with rheumatoid arthritis (RA) experience adverse events because of the characteristics of the disease and the side effects of medications. We investigated the trends of adverse events and mortality associated with disease-modifying antirheumatic drugs (DMARDs). METHODS: We used the Taiwan National Health Insurance Database to enroll patients with incident RA between 2000 and 2017. The 1-year incident rate of gastrointestinal (GI) bleeding and 3-year incident rates of other adverse events and mortality for each calendar-quarter cohort were computed and adjusted using propensity score-based stabilized weights for fair comparisons. Levels and trends of the conventional DMARD era (2000-2002, Phase 1) were compared with those of the TNFi era (2003-2012, Phase 2) and OMA era (2013-2017, Phase 3) by using interrupted time series (ITS) analysis. RESULTS: All patients with RA were prescribed cDMARDs in Phase 1 (2000-2002), and 1%-3% were prescribed either TNFi in phase 2 (2003-2012) or OMAs in phase 3 (2013-2017). The cancer incidence rate was 1.90%, and its mortality rate was 4.19%. After the introduction of TNFi from 2003 to 2012, the main outcomes, except TKA, exhibited a steady or mild decrease in trends. ITS analysis revealed that the slope mildly increased in 2003-2012 compared with that in 2000-2003 by 0.13% for total knee replacement (p = .0322). In 2012-2017 (the OMA era), the events became steady. CONCLUSION: In patients with RA, the introduction of DMARDs was associated with stable adverse events and mortality rates. Moreover, the introduced new treatment for RA exhibited a good safety profile.

End-to-end interstitial fibrosis assessment of kidney biopsies with a machine learning-based model.

BACKGROUND: The extent of interstitial fibrosis in the kidney not only correlates with renal function at the time of biopsy but also predicts future renal outcome. However, its assessment by pathologists lacks good agreement. The aim of this study is to construct a machine learning-based model that enables automatic and reliable assessment of interstitial fibrosis in human kidney biopsies. METHODS: Validated cortex, glomerulus and tubule segmentation algorithms were incorporated into a single model to assess the extent of interstitial fibrosis. The model performances were compared with expert renal pathologists and correlated with patients' renal functional data. RESULTS: Compared with human raters, the model had the best agreement [intraclass correlation coefficient (ICC) 0.90] to the reference in 50 test cases. The model also had a low mean bias and the narrowest 95% limits of agreement. The model was robust against colour variation on images obtained at different times, through different scanners, or from outside institutions with excellent ICCs of 0.92-0.97. The model showed significantly better test-retest reliability (ICC 0.98) than humans (ICC 0.76-0.94) and the amount of interstitial fibrosis inferred by the model strongly correlated with 405 patients' serum creatinine (r = 0.65-0.67) and estimated glomerular filtration rate (r = -0.74 to -0.76). CONCLUSIONS: This study demonstrated that a trained machine learning-based model can faithfully simulate the whole process of interstitial fibrosis assessment, which traditionally can only be carried out by renal pathologists. Our data suggested that such a model may provide more reliable results, thus enabling precision medicine.

Automatic prosthetic-parameter estimation from anteroposterior pelvic radiographs after total hip arthroplasty using deep learning-based keypoint detection.

BACKGROUND: X-ray is a necessary tool for post-total hip arthroplasty (THA) check-ups; however, parameter measurements are time-consuming. We proposed a deep learning tool, BKNet that automates localization of landmarks with parameter measurements. METHODS: About 3072 radiographs from 3021 patients who underwent THA at our institute between 2013 and 2017 were used. We employed BKNet to perform landmark localization with parameter measurements in these radiographs. The performance of BKNet was assessed and compared with that of human observers. RESULTS: The 75-percentile cut-off errors were <0.5 cm in all key points. The Bland-Altman methods show the agreement between the predicted and ground truth parameters. Human and BKNet comparison revealed the model could match the repeatability for 7/10 of the parameters. CONCLUSIONS: The accuracy of BKNet is equivalent to that of human observers, and BKNet was able to perform prosthetic-parameter estimation from keypoint detection with superior cost-effectiveness, repeatability, and timesaving compared to human observers.

A Noninvasive Risk Stratification Tool Build Using an Artificial Intelligence Approach for Colorectal Polyps Based on Annual Checkup Data.

Colorectal cancer is the leading cause of cancer-related deaths worldwide, and early detection has proven to be an effective method for reducing mortality. The machine learning method can be implemented to build a noninvasive stratifying tool that helps identify patients with potential colorectal precancerous lesions (polyps). This study aimed to develop a noninvasive risk-stratified tool for colorectal polyps in asymptomatic, healthy participants. A total of 20,129 consecutive asymptomatic patients who underwent a health checkup between January 2005 and August 2007 were recruited. Positive relationships between noninvasive risk factors, such as age, Helicobacter pylori infection, hypertension, gallbladder polyps/stone, and BMI and colorectal polyps were observed (p < 0.0001), regardless of sex, whereas significant findings were noted in men with tooth disease (p = 0.0053). A risk stratification tool was developed, for colorectal polyps, that considers annual checkup results from noninvasive examinations. For the noninvasive stratified tool, the area under the receiver operating characteristic curve (AUC) of obese females (males) aged <50 years was 91% (83%). In elderly patients (>50 years old), the AUCs of the stratifying tools were >85%. Our results indicate that the risk stratification tool can be built by using random forest and serve as an efficient noninvasive tool to identify patients requiring colonoscopy.

Effect of Blue Light-Filtering Intraocular Lenses on Age-Related Macular Degeneration: A Nationwide Cohort Study With 10-Year Follow-up.

PURPOSE: To determine the incidence rate of age-related macular degeneration (AMD) after cataract surgery and compare the relative incidence of AMD in pseudophakes with blue light-filtering intraocular lenses (BF-IOLs) and non-BF-IOLs. DESIGN: A nationwide cohort study conducted using the Taiwan National Health Insurance Research Database. METHODS: We enrolled 186,591 patients who underwent cataract surgery in both eyes between 2008 and 2013 and monitored them from the index date (the date of first cataract surgery) until AMD, death, loss to follow-up, or December 31, 2017, whichever occurred first. Propensity score matching (PSM) was used to balance the baseline characteristics between the BF-IOL and non-BF-IOL groups. RESULTS: BF-IOLs were implanted in 21,126 patients (11.3%) and non-BF-IOLs were implanted in 165,465 patients (88.7%). Patients in the BF-IOL group tended to be younger, with fewer men, different cataract surgery years, higher income, more nonmanual workers, more patients from urban and suburban areas, and fewer chronic diseases compared with the non-BF-IOL group. With a mean follow-up period of 6.1 years (range, 1-10 years) after cataract surgery, 12,533 and 1655 patients developed non-exudative AMD and exudative AMD, respectively. The incidence rate of non-exudative AMD and exudative AMD (per 1000 person-years) was 9.95 and 1.22 for the BF-IOL group and 11.13 and 1.44 for the non-BF-IOL group, respectively. After PSM, no statistical difference in the incidence rate of nonexudative AMD (hazards ratio, 0.95; 95% CI, 0.88-1.03) and exudative AMD (hazard ratio, 0.96; 95% CI, 0.77-1.18) was observed between the BF-IOL and non-BF-IOL groups. CONCLUSIONS: In Taiwan, the incidence rate of AMD after cataract surgery was 11.59 per 1000 person-years. The use of a BF-IOL for up to 10 years had no apparent advantage over a non-BF-IOL in the incidence of AMD.

Glomerular disease classification and lesion identification by machine learning.

BACKGROUND: Classification of glomerular diseases and identification of glomerular lesions require careful morphological examination by experienced nephropathologists, which is labor-intensive, time-consuming, and prone to interobserver variability. In this regard, recent advance in machine learning-based image analysis is promising. METHODS: We combined Mask Region-based Convolutional Neural Networks (Mask R-CNN) with an additional classification step to build a glomerulus detection model using human kidney biopsy samples. A Long Short-Term Memory (LSTM) recurrent neural network was applied for glomerular disease classification, and another two-stage model using ResNeXt-101 was constructed for glomerular lesion identification in cases of lupus nephritis. RESULTS: The detection model showed state-of-the-art performance on variedly stained slides with F1 scores up to 0.944. The disease classification model showed good accuracies up to 0.940 on recognizing different glomerular diseases based on H&E whole slide images. The lesion identification model demonstrated high discriminating power with area under the receiver operating characteristic curve up to 0.947 for various glomerular lesions. Models showed good generalization on external testing datasets. CONCLUSION: This study is the first-of-its-kind showing how each step of kidney biopsy interpretation carried out by nephropathologists can be captured and simulated by machine learning models. The models were integrated into a whole slide image viewing and annotating platform to enable nephropathologists to review, correct, and confirm the inference results. Further improvement on model performances and incorporating inputs from immunofluorescence, electron microscopy, and clinical data might realize actual clinical use.