Search

OBJECTIVE. Deep learning applications in radiology often suffer from overfitting, limiting generalization to external centers. The objective of this study was to develop a high-quality prostate segmentation model capable of maintaining a high degree of performance across multiple independent datasets using transfer learning and data augmentation. MATERIALS AND METHODS. A retrospective cohort of 648 patients who underwent prostate MRI between February 2015 and November 2018 at a single center was used for training and validation. A deep learning approach combining 2D and 3D architecture was used for training, which incorporated transfer learning. A data augmentation strategy was used that was specific to the deformations, intensity, and alterations in image quality seen on radiology images. Five independent datasets, four of which were from outside centers, were used for testing, which was conducted with and without fine-tuning of the original model. The Dice similarity coefficient was used to evaluate model performance. RESULTS. When prostate segmentation models utilizing transfer learning were applied to the internal validation cohort, the mean Dice similarity coefficient was 93.1 for whole prostate and 89.0 for transition zone segmentations. When the models were applied to multiple test set cohorts, the improvement in performance achieved using data augmentation alone was 2.2% for the whole prostate models and 3.0% for the transition zone segmentation models. However, the best test-set results were obtained with models fine-tuned on test center data with mean Dice similarity coefficients of 91.5 for whole prostate segmentation and 89.7 for transition zone segmentation. CONCLUSION. Transfer learning allowed for the development of a high-performing prostate segmentation model, and data augmentation and fine-tuning approaches improved performance of a prostate segmentation model when applied to datasets from external centers.

OBJECTIVE. The purpose of this study was to evaluate in a multicenter dataset the performance of an artificial intelligence (AI) detection system with attention mapping compared with multiparametric MRI (mpMRI) interpretation in the detection of prostate cancer. MATERIALS AND METHODS. MRI examinations from five institutions were included in this study and were evaluated by nine readers. In the first round, readers evaluated mpMRI studies using the Prostate Imaging Reporting and Data System version 2. After 4 weeks, images were again presented to readers along with the AI-based detection system output. Readers accepted or rejected lesions within four AI-generated attention map boxes. Additional lesions outside of boxes were excluded from detection and categorization. The performances of readers using the mpMRI-only and AI-assisted approaches were compared. RESULTS. The study population included 152 case patients and 84 control patients with 274 pathologically proven cancer lesions. The lesion-based AUC was 74.9% for MRI and 77.5% for AI with no significant difference ( p = 0.095). The sensitivity for overall detection of cancer lesions was higher for AI than for mpMRI but did not reach statistical significance (57.4% vs 53.6%, p = 0.073). However, for transition zone lesions, sensitivity was higher for AI than for MRI (61.8% vs 50.8%, p = 0.001). Reading time was longer for AI than for MRI (4.66 vs 4.03 minutes, p < 0.001). There was moderate interreader agreement for AI and MRI with no significant difference (58.7% vs 58.5%, p = 0.966). CONCLUSION. Overall sensitivity was only minimally improved by use of the AI system. Significant improvement was achieved, however, in the detection of transition zone lesions with use of the AI system at the cost of a mean of 40 seconds of additional reading time.

Objective: The purpose of this study is to evaluate the role of MRI for characterization of high-risk breast lesions diagnosed at imaging-guided needle biopsy., Materials and Methods: In this retrospective analysis of 220 patients, 227 high-risk lesions (94 papillomas, 64 radial sclerosing lesions, 46 lobular neoplasias, and 23 atypical ductal hyperplasias) found at 11-gauge vacuum-assisted or 14-gauge needle biopsy were studied with dynamic MRI (time resolution, 84 or 88 seconds; gadopentetate dimeglumine or gadobenate dimeglumine, 0.1 mmol/kg). When lesions showed contrast enhancement on subtracted images, they were considered suspicious for malignancy. The reference standard was histopathologic examination after surgical excision in 190 of 227 (84%) lesions and negative follow-up (≥ 24 months) in 37 of 227 (16%) lesions. Predictive values and likelihood ratios were calculated., Results: Of 227 lesions, 155 (68%) were contrast enhancing and 72 (32%) were not. Of 155 contrast-enhancing lesions, 28 (18%) were upgraded to malignancy after surgical excision (nine papillomas, one radial sclerosing lesion, 11 lobular neoplasias, and seven atypical ductal hyperplasias); there were 11 invasive carcinomas and 17 ductal carcinomas in situ, four of the latter being G3. Of 72 non-contrast-enhancing lesions, two (3%) were upgraded to malignancy after surgical excision (one radial sclerosing lesion and one lobular neoplasia), both of which were G1 ductal carcinoma in situ. Cancer probability was significantly higher for contrast-enhancing (18%) than for non-contrast-enhancing (3%) lesions (p = 0.001) and for nonmasslike (43%) than for masslike (14%) lesions (p = 0.005). The positive predictive value was 18% (28/155; 95% CI, 13-24%), the negative predictive value was 97% (70/72; 95% CI, 94-99%), the positive likelihood ratio was 1.448 (95% CI, 1.172-1.788), and the negative likelihood ratio was 0.188 (95% CI, 0.152-0.232)., Conclusion: The absence of enhancement at dynamic MRI allowed reliable exclusion of invasive cancers among high-risk lesions diagnosed at needle biopsy.

Objective: The purpose of this study was to investigate whether breast MRI can be used to rule out malignancy in patients with high-risk lesions diagnosed at imaging-guided core needle biopsy., Subjects and Methods: The subjects were women consecutively registered between October 2004 and April 2010 who had high-risk lesions diagnosed at mammographically or sonographically guided core needle biopsy and subsequently underwent MRI and surgical excision. MR images were reviewed by two experienced breast radiologists. Lesions assessed as BI-RADS category 1-3 were considered negative for malignancy, and BI-RADS 4 and 5 lesions were considered malignant. Histologic findings at surgical excision were the reference standard. The sensitivity, specificity, and positive and negative predictive values of MRI in the detection of associated malignancy were calculated for the entire set of lesions and for each histologic subtype., Results: The final sample consisted of 169 high-risk lesions in 166 patients. At MRI analysis, 116 (68.6%) lesions were considered negative for malignancy, and the other 53 (31.4%) malignant. At surgical excision, 22 malignant lesions were found. The overall sensitivity, specificity, and positive and negative predictive values of MRI were 72.7% (16/22), 74.8% (110/147), 30.2% (16/53), and 94.8% (110/116). The negative predictive values for papilloma, radial scar, lobular neoplasia, and atypical ductal hyperplasia were 97.4% (38/39), 97.6% (41/42), 88.0% (22/25), and 90.0% (9/10)., Conclusion: Patients with high-risk lesions associated with the lowest likelihood of malignancy (papilloma and radial scar) and without suspicious MRI findings can safely undergo follow-up instead of surgery. Because of the low negative predictive value, however, MRI is not helpful in cases of lobular neoplasia and atypical ductal hyperplasia, and all these lesions should be excised.

Objective: The purposes of our study were, first, to evaluate the frequency, clinical presentation, and associated imaging findings of malignant breast lesions presenting as hyperechoic nodules in a large series of consecutive sonographically guided core needle biopsies (CNBs) and, second, to investigate sonographic features that are able to predict malignancy in hyperechoic breast lesions., Materials and Methods: The radiologic and pathologic records for 4511 consecutive sonographically guided CNBs were retrospectively reviewed. Hyperechoic lesions were identified, and clinical notes and related mammography or MRI reports were reviewed. The sonographic images were evaluated according to the BI-RADS lexicon by two experienced breast radiologists. Surgical pathology results and follow-up served as the reference standard for lesions diagnosed as malignant or high-risk and benign at CNB, respectively. The frequency of hyperechoic carcinomas among all carcinomas was calculated. Differences in sonographic appearance between hyperechoic benign and malignant lesions were evaluated using the chi-square test or the Fisher exact test., Results: Of all biopsied lesions, 25 (0.6%) were hyperechoic. Among the 1849 malignant lesions, nine (0.4%) were hyperechoic. The remaining 16 were benign. None of the hyperechoic malignancies was a "purely" sonographic lesion, because all were palpable, mammographically visible, or detectable on breast MRI. Malignant lesions were more likely than benign lesions to have noncircumscribed margins (9/9 vs 7/16; p = 0.008) and nonparallel orientation (6/9 vs 1/16; p = 0.003)., Conclusion: When encountering a hyperechoic nodule, malignant nature cannot be excluded. Suspicious sonographic signs and correlation with other imaging techniques may help avoid misdiagnosis.

Objective: The purposes of our study were to evaluate the surgical outcome of cases of radial scar without atypia diagnosed at imaging-guided percutaneous needle biopsy and to determine whether the mammographic and sonographic features are able to predict which lesions will be upgraded to malignancy at surgical excision., Materials and Methods: The records of 4,458 consecutive imaging-guided biopsies were retrospectively reviewed. Surgical excision results were available in 62 cases in which radial scar was the highest-risk lesion at stereotactically guided or sonographically guided biopsy. The mammographic and sonographic images and surgical findings were reviewed. The underestimation rate of malignancy of percutaneous biopsy was calculated. Differences in mammographic and sonographic appearances between radial scars with and without associated malignancy were evaluated using the Fisher's exact test., Results: The percutaneous malignancy underestimation rate was 8% (5/62): 9% (4/43) for sonography guided 14-gauge biopsies and 5% (1/19) for stereotactically guided 11-gauge vacuum-assisted biopsies (p = 1.000). Mammographic and sonographic appearances were not significantly different between radial scars with and those without associated malignancy., Conclusion: A percutaneous diagnosis of a radial scar does not exclude associated malignancy at surgical excision. Mammographic and sonographic features of a lesion diagnosed as a radial scar at percutaneous imaging-guided biopsy do not predict which lesions will have associated malignancy at surgery. Therefore, all patients with percutaneous diagnosis of a radial scar should undergo surgical excision regardless of mammographic and sonographic appearances, until further criteria can be determined.