Biopsy-free AI-aided precision MRI assessment in prediction of prostate cancer biochemical recurrence.

BACKGROUND: To investigate the predictive ability of high-throughput MRI with deep survival networks for biochemical recurrence (BCR) of prostate cancer (PCa) after prostatectomy. METHODS: Clinical-MRI and histopathologic data of 579 (train/test, 463/116) PCa patients were retrospectively collected. The deep survival network (iBCR-Net) is based on stepwise processing operations, which first built an MRI radiomics signature (RadS) for BCR, and predicted the T3 stage and lymph node metastasis (LN+) of tumour using two predefined AI models. Subsequently, clinical, imaging and histopathological variables were integrated into iBCR-Net for BCR prediction. RESULTS: RadS, derived from 2554 MRI features, was identified as an independent predictor of BCR. Two predefined AI models achieved an accuracy of 82.6% and 78.4% in staging T3 and LN+. The iBCR-Net, when expressed as a presurgical model by integrating RadS, AI-diagnosed T3 stage and PSA, can match a state-of-the-art histopathological model (C-index, 0.81 to 0.83 vs 0.79 to 0.81, p > 0.05); and has maximally 5.16-fold, 12.8-fold, and 2.09-fold (p < 0.05) benefit to conventional D'Amico score, the Cancer of the Prostate Risk Assessment (CAPRA) score and the CAPRA Postsurgical score. CONCLUSIONS: AI-aided iBCR-Net using high-throughput MRI can predict PCa BCR accurately and thus may provide an alternative to the conventional method for PCa risk stratification.

Hospital crowdedness evaluation and in-hospital resource allocation based on image recognition technology.

How to allocate the existing medical resources reasonably, alleviate hospital congestion and improve the patient experience are problems faced by all hospitals. At present, the combination of artificial intelligence and the medical field is mainly in the field of disease diagnosis, but lacks successful application in medical management. We distinguish each area of the emergency department by the division of medical links. In the spatial dimension, in this study, the waitlist number in real-time is got by processing videos using image recognition via a convolutional neural network. The congestion rate based on psychology and architecture is defined for measuring crowdedness. In the time dimension, diagnosis time and time-consuming after diagnosis are calculated from visit records. Factors related to congestion are analyzed. A total of 4717 visit records from the emergency department and 1130 videos from five areas are collected in the study. Of these, the waiting list of the pediatric waiting area is the largest, including 10,436 (person-time) people, and its average congestion rate is 2.75, which is the highest in all areas. The utilization rate of pharmacy is low, with an average of only 3.8 people using it at the one time. Its average congestion rate is only 0.16, and there is obvious space waste. It has been found that the length of diagnosis time and the length of time after diagnosis are related to age, the number of diagnoses and disease type. The most common disease type comes from respiratory problems, accounting for 54.3%. This emergency department has congestion and waste of medical resources. People can use artificial intelligence to investigate the congestion in hospitals effectively. Using artificial intelligence methods and traditional statistics methods can lead to better research on healthcare resource allocation issues in hospitals.

High-throughput precision MRI assessment with integrated stack-ensemble deep learning can enhance the preoperative prediction of prostate cancer Gleason grade.

BACKGROUND: To develop and test a Prostate Imaging Stratification Risk (PRISK) tool for precisely assessing the International Society of Urological Pathology Gleason grade (ISUP-GG) of prostate cancer (PCa). METHODS: This study included 1442 patients with prostate biopsy from two centres (training, n = 672; internal test, n = 231 and external test, n = 539). PRISK is designed to classify ISUP-GG 0 (benign), ISUP-GG 1, ISUP-GG 2, ISUP-GG 3 and ISUP GG 4/5. Clinical indicators and high-throughput MRI features of PCa were integrated and modelled with hybrid stacked-ensemble learning algorithms. RESULTS: PRISK achieved a macro area-under-curve of 0.783, 0.798 and 0.762 for the classification of ISUP-GGs in training, internal and external test data. Permitting error ±1 in grading ISUP-GGs, the overall accuracy of PRISK is nearly comparable to invasive biopsy (train: 85.1% vs 88.7%; internal test: 85.1% vs 90.4%; external test: 90.4% vs 94.2%). PSA ≥ 20 ng/ml (odds ratio [OR], 1.58; p = 0.001) and PRISK ≥ GG 3 (OR, 1.45; p = 0.005) were two independent predictors of biochemical recurrence (BCR)-free survival, with a C-index of 0.76 (95% CI, 0.73-0.79) for BCR-free survival prediction. CONCLUSIONS: PRISK might offer a potential alternative to non-invasively assess ISUP-GG of PCa.

Optimized MRI Assessment for Clinically Significant Prostate Cancer: A STARD-Compliant Two-Center Study.

BACKGROUND: There is a requirement for a personalized strategy to make MRI more accessible to men with suspicion of clinically significant prostate cancer (CSPC). PURPOSE: To evaluate an optimized (Op)-MRI compared with biparametric (Bp)-MRI and multiparametric (Mp)-MRI for the diagnosis of CSPC. STUDY TYPE: Two-center, retrospective. SUBJECTS: A total of 346 patients from center 1 and 292 patients from center 2. FIELD STRENGTH/SEQUENCE: 3.0T scanners, T2 -weighted imaging (T2 WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced (DCE) imaging. ASSESSMENT: Four radiologists interpreted the Bp-MRI (T2 WI and DWI) and Mp-MRI (T2 WI, DWI, and DCE) independently according to the Prostate Imaging Reporting and Data System (PI-RADS). For Op-MRI, two radiologists used an adjusted decision rule on Bp-MRI-assessed PI-RADS 3 lesions by determining early enhancement of DCE. Pathologies at biopsy and/or prostatectomy specimens were used as standard references. STATISTICAL TESTS: Performance was assessed using receiver operating characteristic (ROC) curves. Kappa statistics were used to assess interobserver variability. RESULTS: Interreader agreement was excellent for all three MRI assessments (all κ values >0.80). Op-MRI had comparable sensitivity (senior/junior: 90.9% [261/287] / 91.6% [263/287]) and higher specificity (78.1% [274/351] /74.4% [261/351]) compared with Mp-MRI (sensitivity: 92.3% [265/287] / 93.7% [269/287]; specificity: 67.8% [238/351] / 68.1% [239/351]) and Bp-MRI (sensitivity: 91.6% [263/287] / 93.4% [268/287]; specificity: 71.2% [250/351] / 70.1% [246/351]) for the diagnosis of CSPC. Compared to Mp-MRI, Op-MRI spared biopsy in 80.7% (515/638) of DCE scans with similar performance accuracy. Compared to Bp-MRI, Op-MRI downgraded 25.2% (31/123) of lesions at a cost of missing 6.5% (3/46) of malignancies, and upgraded 45.5% (56/123) of lesions with a positive predictive value of 62.5% (35/56) in 123 equivocal findings. DATA CONCLUSION: The Op-MRI, using an adjusted PI-RADS decision rule, did not compromise diagnostic accuracy with sparing biopsy in 80.7% of DCE scans compared to Mp-MRI, and outperformed Bp-MRI by regrading PI-RADS lesions. LEVEL OF EVIDENCE: 4 TECHNICAL EFFICACY STAGE: 2.

Performance of a machine learning-based decision model to help clinicians decide the extent of lymphadenectomy (D1 vs. D2) in gastric cancer before surgical resection.

BACKGROUND: Controversy still exists on the optimal surgical resection for potentially curable gastric cancer (GC). Use of radiologic evaluation and machine learning algorithms might predict extent of lymphadenectomy to limit unnecessary surgical treatment. We purposed to design a machine learning-based clinical decision-support model for predicting extent of lymphadenectomy (D1 vs. D2) in local advanced GC. METHODS: Clinicoradiologic features available from routine clinical assignments in 557 patients with GCs were retrospectively interpreted by an expert panel blinded to all histopathologic information. All patients underwent surgery using standard D2 resection. Decision models were developed with a logistic regression (LR), support vector machine (SVM) and auto-encoder (AE) algorithm in 371 training and tested in 186 test data, respectively. The primary end point was to measure diagnostic performance of decision model and a Japanese gastric cancer treatment guideline version 4th (JPN 4th) criteria for discriminate D1 (pT1 + pN0) versus D2 (≥ pT1 + ≥ pN1) lymphadenectomy. RESULTS: The decision model with AE analysis produced highest area under ROC curve (train: 0.965, 95% confidence interval (CI) 0.948-0.978; test: 0.946, 95% CI 0.925-0.978), followed by SVM (train: 0.925, 95% CI 0.902-0.944; test: 0.942, 95% CI 0.922-0.973) and LR (train: 0.886, 95% CI 0.858-0.910; test: 0.891, 95% CI 0.891-0.952). By this improvement, overtreatment was reduced from 21.7% (121/557) by treat-all pattern, to 15.1% (84/557) by JPN 4th criteria, and to 0.7-0.9% (4-5/557) by the new approach. CONCLUSIONS: The decision model with machine learning analysis demonstrates high accuracy for identifying patients who are candidates for D1 versus D2 resection. Its approximate 14-20% improvements in overtreatment compared to treat-all pattern and JPN 4th criteria potentially increase the number of patients with local advanced GCs who can safely avoid unnecessary lymphadenectomy.

Volumetric analysis of intravoxel incoherent motion imaging for assessment of solitary pulmonary lesions.

Background Differentiating between malignant and benign solitary pulmonary lesions (SPLs) is challenging. Purpose To determine diagnostic performance of intravoxel incoherent motion-based diffusion-weighted imaging (DW-IVIM) in distinguishing malignant from benign SPLs, using histogram analysis derived whole-tumor and single-section region of interest (ROI). Material and Methods This retrospective study received institutional review board approval. A total of 129 patients with diagnosed SPLs underwent DW-IVIM and apparent diffusion coefficient (ADC). ADC, slow diffusion coefficient (D), fast diffusion coefficient (D*), and perfusion fraction (f) were calculated separately by outlining whole-tumor and single-section ROI. Inter-observer reliability was assessed by inter-class correlation coefficient (ICC). ADC and DW-IVIM parameters were analyzed using independent-sample T-test. Receiver operating characteristic (ROC) analysis was constructed to determine diagnostic performance. Multiple logistic regression was performed to identify independent factors associated with malignant SPLs. Results There were 48 benign SPLs found in 35 patients and 94 patients with lung cancer (LC). ICC for whole-tumor ROI (range, 0.89-0.95) was higher than that for single-section ROI (range, 0.61-0.71). Mean ADC and D were significantly lower in the malignant group. ADC and D 10th showed significantly higher AUC values than did mean ADC and D. D showed significantly higher diagnostic accuracy in mean, 10th, and 25th percentiles than ADC values (all Ps < 0.05). D 10th was found to be an independent factor in discriminating LCs with an odds ratio of -1.217. Conclusion Volumetric analysis had higher reproducibility and diagnostic accuracy than did single-section. Further, compared to ADC, D value differentiated benign SPLs from LCs more accurately.

Lung adenocarcinoma: Assessment of epidermal growth factor receptor mutation status based on extended models of diffusion-weighted image.

PURPOSE: To evaluate the diagnostic performance of extended models of diffusion-weighted (DW) imaging to help differentiate the epidermal growth factor receptor (EGFR) mutation status in stage IIIA-IV lung adenocarcinoma. MATERIALS AND METHODS: This retrospective study had institutional research board approval and was HIPAA compliant. Preoperative extended DW imaging including intravoxel incoherent motion (IVIM) and diffusional kurtosis imaging (DKI) 3 Tesla MRI were retrospectively evaluated in 53 patients with pathologically confirmed non-early stage (IIIA-IV) lung adenocarcinoma. EGFR mutationsat exons 18-21 were determined by using polymerase chain reaction-based ARMS. Quantitative parameters (mean, kurtosis, skewness, 10th and 90th percentiles) of IVIM (true-diffusion coefficient D, pseudo-diffusion coefficient D*, and perfusion fraction f) and DKI (kurtosis value Kapp, kurtosis corrected diffusion coefficient Dapp) were calculated by outlining entire-volume histogram analysis. Receiver operating characteristic analysis was constructed to determine the diagnostic performance of each parameter. Multivariate logistic regression was used to differentiate the probability of EGFR mutation status. RESULTS: Twenty-four of 53 patients with lung adenocarcinoma were EGFR mutations, which occurred most often in acinar (10 of 13 [76.9%]) and papillary predominant tumors (9 of 13 [69.2%]). Patients with EGFR mutation showed significant higher 10th percentile of D, lower D* value in terms of kurtosis, and lower Kapp value in terms of mean, skewness, 10th and 90th percentiles (all P values < 0.05). The 90th Kapp showed significantly higher sensitivity (97%; P < 0.05) and Az (0.817; P < 0.05) value. Multivariate logistic regression showed 90th Kapp was a independent factor for determining EGFR mutation with odds ratio -1.657. CONCLUSION: Multiple IVIM and DKI parameters, especially the histogram 90th Kapp value, helped differentiate EGFR mutation status in stage IIIA-IV lung adenocarcinoma. LEVEL OF EVIDENCE: 3 Technical Efficacy: Stage 2 J. MAGN. RESON. IMAGING 2017;46:281-289.

The Diagnostic Performance of Coronary CT Angiography for the Assessment of Coronary Stenosis in Calcified Plaque.

PURPOSE: To prospectively evaluate the diagnostic performance of coronary CT angiography (CCTA) for the assessment of coronary stenosis in a calcified plaque, by using conventional coronary angiography (CAG) as a standard reference. MATERIALS AND METHODS: Eight hundred and ninety-four patients were known to have or have been suspicious of having coronary artery disease, underwent CCTA and conventional coronary angiography (CAG). All the images acquired were assessed. The calcified plaque in CCTA was classified into four types (I-IV) according to the ratio of calcified plaque volume to vessel circumference (RVTC). Overall diagnostic accuracy was made under receiver operating characteristic curve (AUC) analysis. CAG was used as the standard reference. RESULTS: A total of 12845 segments were evaluated in 894 patients, among which 4955 calcified plaques were detected on 3645(28.4%) segments by CCTA. The overall AUC, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were 0.939, 97.8%, 90.1%, 71.2% and 99.4%, respectively. In type I-II calcification, CCTA had high diagnostic performance in AUC (type I: 0.983; type II: 0.976), sensitivity (96.7%; 98.1%), specificity (99.8%; 97.0%), PPV (95.7%; 90.1%), NPV (99.8%; 99.5%) and accuracy (99.6%; 97.3%). In type III-IV calcification, CCTA has high performance in sensitivity (type III: 97.6%; type IV: 97.9%) and NPV (98.3%; 98.7%), moderate performance in AUC (0.877; 0.829), while remarkable decrease in specificity (78.7%; 67.9%), PPV (71.0%; 56.2%) and accuracy (84.9%; 76.8%). CONCLUSION: CCTA has highest accuracy in diagnosing the coronary artery stenosis of type I-II calcified plaques, but has a significant decrease in specificity, PPV and accuracy in type III-IV calcified plaque.