Background Clinically significant prostate cancer (PCa) diagnosis at MRI requires accurate and efficient radiologic interpretation. Although artificial intelligence may assist in this task, lack of transparency has limited clinical translation. Purpose To develop an explainable artificial intelligence (XAI) model for clinically significant PCa diagnosis at biparametric MRI using Prostate Imaging Reporting and Data System (PI-RADS) features for classification justification. Materials and Methods This retrospective study included consecutive patients with histopathologic analysis-proven prostatic lesions who underwent biparametric MRI and biopsy between January 2012 and December 2017. After image annotation by two radiologists, a deep learning model was trained to detect the index lesion; classify PCa, clinically significant PCa (Gleason score ≥ 7), and benign lesions (eg, prostatitis); and justify classifications using PI-RADS features. Lesion- and patient-based performance were assessed using fivefold cross validation and areas under the receiver operating characteristic curve. Clinical feasibility was tested in a multireader study and by using the external PROSTATEx data set. Statistical evaluation of the multireader study included Mann-Whitney U and exact Fisher-Yates test. Results Overall, 1224 men (median age, 67 years; IQR, 62-73 years) had 3260 prostatic lesions (372 lesions with Gleason score of 6; 743 lesions with Gleason score of ≥ 7; 2145 benign lesions). XAI reliably detected clinically significant PCa in internal (area under the receiver operating characteristic curve, 0.89) and external test sets (area under the receiver operating characteristic curve, 0.87) with a sensitivity of 93% (95% CI: 87, 98) and an average of one false-positive finding per patient. Accuracy of the visual and textual explanations of XAI classifications was 80% (1080 of 1352), confirmed by experts. XAI-assisted readings improved the confidence (4.1 vs 3.4 on a five-point Likert scale; P = .007) of nonexperts in assessing PI-RADS 3 lesions, reducing reading time by 58 seconds (P = .009). Conclusion The explainable AI model reliably detected and classified clinically significant prostate cancer and improved the confidence and reading time of nonexperts while providing visual and textual explanations using well-established imaging features. © RSNA, 2023 Supplemental material is available for this article. See also the editorial by Chapiro in this issue.

Background: To assess the impact of body composition imaging biomarkers in computed tomography (CT) on the perioperative morbidity and survival after surgery of patients with esophageal cancer (EC). Methods: Eighty-five patients who underwent esophagectomy for locally advanced EC after neoadjuvant therapy between 2014 and 2019 were retrospectively enrolled. Pre- and postoperative CT scans were used to assess the body composition imaging biomarkers (visceral (VAT) and subcutaneous adipose tissue (SAT) areas, psoas muscle area (PMA) and volume (PMV), total abdominal muscle area (TAMA)). Sarcopenia was defined as lumbar skeletal muscle index (LSMI) ≤38.5 cm²/m² in women and ≤52.4 cm²/m² in men. Patients with a body mass index (BMI) of ≥30 were considered obese. These imaging biomarkers were correlated with major complications, anastomotic leakage, postoperative pneumonia, duration of postoperative hospitalization, disease-free survival (DFS), and overall survival (OS). Results: Preoperatively, sarcopenia was identified in 58 patients (68.2%), and sarcopenic obesity was present in 7 patients (8.2%). Sarcopenic patients were found to have an elevated risk for the occurrence of major complications (OR: 2.587, p = 0.048) and prolonged hospitalization (32 d vs. 19 d, p = 0.040). Patients with sarcopenic obesity had a significantly higher risk for postoperative pneumonia (OR: 6.364 p = 0.018) and a longer postoperative hospital stay (71 d vs. 24 d, p = 0.021). Neither sarcopenia nor sarcopenic obesity was an independent risk factor for the occurrence of anastomotic leakage (p > 0.05). Low preoperative muscle biomarkers (PMA and PMV) and their decrease (ΔPMV and ΔTAMA) during the follow-up period significantly correlated with shorter DFS and OS (p = 0.005 to 0.048). Conclusion: CT body composition imaging biomarkers can identify high-risk patients with locally advanced esophageal cancer undergoing surgery. Sarcopenic patients have a higher risk of major complications, and patients with sarcopenic obesity are more prone to postoperative pneumonia. Sarcopenia and sarcopenic obesity are both subsequently associated with a prolonged hospitalization. Low preoperative muscle mass and its decrease during the postoperative follow-up are associated with lower DFS and OS.

Background: Patients with Marfan syndrome are at risk for aortic enlargement and are routinely monitored by computed tomography (CT) imaging. The purpose of this study is to analyse body composition using artificial intelligence (AI)‐based tissue segmentation in patients with Marfan syndrome in order to identify possible predictors of progressive aortic enlargement.Methods: In this study, the body composition of 25 patients aged ≤50 years with Marfan syndrome and no prior aortic repair was analysed at the third lumbar vertebra (L3) level from a retrospective dataset using an AI‐based software tool (Visage Imaging). All patients underwent electrocardiography‐triggered CT of the aorta twice within 2 years for suspected progression of aortic disease, suspected dissection, and/or pre‐operative evaluation. Progression of aortic enlargement was defined as an increase in diameter at the aortic sinus or the ascending aorta of at least 2 mm. Patients meeting this definition were assigned to the ‘progressive aortic enlargement’ group (proAE group) and patients with stable diameters to the ‘stable aortic enlargement’ group (staAE group). Statistical analysis was performed using the Mann–Whitney U test. Two possible body composition predictors of aortic enlargement—skeletal muscle density (SMD) and psoas muscle index (PMI)—were analysed further using multivariant logistic regression analysis. Aortic enlargement was defined as the dependent variant, whereas PMI, SMD, age, sex, body mass index (BMI), beta blocker medication, and time interval between CT scans were defined as independent variants.Results: There were 13 patients in the proAE group and 12 patients in the staAE group. AI‐based automated analysis of body composition at L3 revealed a significantly increased SMD measured in Hounsfield units (HUs) in patients with aortic enlargement (proAE group: 50.0 ± 8.6 HU vs. staAE group: 39.0 ± 15.0 HU; P = 0.03). PMI also trended towards higher values in the proAE group (proAE group: 6.8 ± 2.3 vs. staAE group: 5.6 ± 1.3; P = 0.19). Multivariate logistic regression revealed significant prediction of aortic enlargement for SMD (P = 0.05) and PMI (P = 0.04).Conclusions: Artificial intelligence‐based analysis of body composition at L3 in Marfan patients is feasible and easily available from CT angiography. Analysis of body composition at L3 revealed significantly higher SMD in patients with progressive aortic enlargement. PMI and SMD significantly predicted aortic enlargement in these patients. Using body composition as a predictor of progressive aortic enlargement may contribute information for risk stratification regarding follow‐up intervals and the need for aortic repair.

Background: To externally evaluate the first picture archiving communications system (PACS)‐integrated artificial intelligence (AI)‐based workflow, trained to automatically detect a predefined computed tomography (CT) slice at the third lumbar vertebra (L3) and automatically perform complete image segmentation for analysis of CT body composition and to compare its performance with that of an established semi‐automatic segmentation tool regarding speed and accuracy of tissue area calculation.Methods: For fully automatic analysis of body composition with L3 recognition, U‐Nets were trained (Visage) and compared with a conventional image segmentation software (TomoVision). Tissue was differentiated into psoas muscle, skeletal muscle, visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT). Mid‐L3 level images from randomly selected DICOM slice files of 20 CT scans acquired with various imaging protocols were segmented with both methods.Results: Success rate of AI‐based L3 recognition was 100%. Compared with semi‐automatic, fully automatic AI‐based image segmentation yielded relative differences of 0.22% and 0.16% for skeletal muscle, 0.47% and 0.49% for psoas muscle, 0.42% and 0.42% for VAT and 0.18% and 0.18% for SAT. AI‐based fully automatic segmentation was significantly faster than semi‐automatic segmentation (3 ± 0 s vs. 170 ± 40 s, P < 0.001, for User 1 and 152 ± 40 s, P < 0.001, for User 2).Conclusion: Rapid fully automatic AI‐based, PACS‐integrated assessment of body composition yields identical results without transfer of critical patient data. Additional metabolic information can be inserted into the patient's image report and offered to the referring clinicians.

Background: Magnetic resonance imaging (MRI) is used to detect the prostate index lesion before targeted biopsy. However, the number of biopsy cores that should be obtained from the index lesion is unclear. The aim of this study is to analyze how many MRI-targeted biopsy cores are needed to establish the most relevant histopathologic diagnosis of the index lesion and to build a prediction model. Methods: We retrospectively included 451 patients who underwent 10-core systematic prostate biopsy and MRI-targeted biopsy with sampling of at least three cores from the index lesion. A total of 1587 biopsy cores were analyzed. The core sampling sequence was recorded, and the first biopsy core detecting the most relevant histopathologic diagnosis was identified. In a subgroup of 261 patients in whom exactly three MRI-targeted biopsy cores were obtained from the index lesion, we generated a prediction model. A nonparametric Bayes classifier was trained using the PI-RADS score, prostate-specific antigen (PSA) density, lesion size, zone, and location as covariates. Results: The most relevant histopathologic diagnosis of the index lesion was detected by the first biopsy core in 331 cases (73%), by the second in 66 cases (15%), and by the third in 39 cases (9%), by the fourth in 13 cases (3%), and by the fifth in two cases (<1%). The Bayes classifier correctly predicted which biopsy core yielded the most relevant histopathologic diagnosis in 79% of the subjects. PI-RADS score, PSA density, lesion size, zone, and location did not independently influence the prediction model. Conclusion: The most relevant histopathologic diagnosis of the index lesion was made on the basis of three MRI-targeted biopsy cores in 97% of patients. Our classifier can help in predicting the first MRI-targeted biopsy core revealing the most relevant histopathologic diagnosis; however, at least three MRI-targeted biopsy cores should be obtained regardless of the preinterventionally assessed covariates.