Development and Validation of a Convolutional Neural Network for Automated Detection of Scaphoid Fractures on Conventional Radiographs.

PURPOSE: To compare the performance of a convolutional neural network (CNN) to that of 11 radiologists in detecting scaphoid bone fractures on conventional radiographs of the hand, wrist, and scaphoid. MATERIALS AND METHODS: At two hospitals (hospitals A and B), three datasets consisting of conventional hand, wrist, and scaphoid radiographs were retrospectively retrieved: a dataset of 1039 radiographs (775 patients [mean age, 48 years ± 23 {standard deviation}; 505 female patients], period: 2017-2019, hospitals A and B) for developing a scaphoid segmentation CNN, a dataset of 3000 radiographs (1846 patients [mean age, 42 years ± 22; 937 female patients], period: 2003-2019, hospital B) for developing a scaphoid fracture detection CNN, and a dataset of 190 radiographs (190 patients [mean age, 43 years ± 20; 77 female patients], period: 2011-2020, hospital A) for testing the complete fracture detection system. Both CNNs were applied consecutively: The segmentation CNN localized the scaphoid and then passed the relevant region to the detection CNN for fracture detection. In an observer study, the performance of the system was compared with that of 11 radiologists. Evaluation metrics included the Dice similarity coefficient (DSC), Hausdorff distance (HD), sensitivity, specificity, positive predictive value (PPV), and area under the receiver operating characteristic curve (AUC). RESULTS: The segmentation CNN achieved a DSC of 97.4% ± 1.4 with an HD of 1.31 mm ± 1.03. The detection CNN had sensitivity of 78% (95% CI: 70, 86), specificity of 84% (95% CI: 77, 92), PPV of 83% (95% CI: 77, 90), and AUC of 0.87 (95% CI: 0.81, 0.91). There was no difference between the AUC of the CNN and that of the radiologists (0.87 [95% CI: 0.81, 0.91] vs 0.83 [radiologist range: 0.79-0.85]; P = .09). CONCLUSION: The developed CNN achieved radiologist-level performance in detecting scaphoid bone fractures on conventional radiographs of the hand, wrist, and scaphoid.Keywords: Convolutional Neural Network (CNN), Deep Learning Algorithms, Machine Learning Algorithms, Feature Detection-Vision-Application Domain, Computer-Aided DiagnosisSee also the commentary by Li and Torriani in this issue.Supplemental material is available for this article.©RSNA, 2021.

Comparison of the Heel Enthesitis MRI Scoring System (HEMRIS) with clinical enthesitis and local metabolic activity on PET-CT.

OBJECTIVE: To compare the Heel Enthesitis MRI Scoring model (HEMRIS) with clinical and PET/CT outcomes in patients with cutaneous psoriasis (Pso), psoriatic arthritis (PsA) or ankylosing spondylitis (AS). METHODS: This prospective, observational study included 38 patients with Pso, PsA and AS. Patients were included regardless of presence or absence of clinical heel enthesitis. MRI-scans of both ankles and a whole-body 18F-FDG PET/CT were acquired. MRIs were assessed for enthesitis by two independent and blinded observers according to the HEMRIS. A physician, blinded for imaging results, performed clinical evaluations of enthesitis at the Achilles tendon and plantar fascia. RESULTS: In total, 146 entheses were scored according to the HEMRIS and clinically assessed for enthesitis (6 entheses were clinically affected). In Achilles tendons with clinical enthesitis, the HEMRIS structural damage score was significantly higher, compared to Achilles tendons without clinical enthesitis (respective median scores 1.0 and 0.5; p=0.04). In clinically unaffected entheses, HEMRIS abnormalities occurred in 44/70 (63%) of Achilles tendons and in 23/70 (33%) of plantar fascia. At the Achilles tendon, local metabolic activity measured on PET/CT was weakly associated with the structural (rs=0.25, p=0.03) and total HEMRIS (rs=0.26, p=0.03). CONCLUSION: This study revealed a high prevalence of subclinical HEMRIS abnormalities and discrepancy between HEMRIS and clinical and PET/CT findings. This may suggest that the HEMRIS is a sensitive method for detection of inflammatory and structural disease of enthesitis at the Achilles tendon and plantar fascia, although the clinical significance of these MRI findings remains to be determined in longitudinal studies.

Clinical utility of ultrasound-needle biopsy for preoperative staging of the axilla in invasive breast cancer.

Sentinel node biopsy (SNB) has largely replaced axillary lymph node dissection (ALND) as the standard-of-care for nodal staging in invasive breast cancer. Preoperative imaging-based staging of the axilla using ultrasound with selective ultrasound-guided needle biopsy (UNB) is moderately-sensitive and identifies approximately 50% of patients (pooled estimate from meta-analysis 50%; 95% confidence interval=43%-57%) with axillary nodal metastases prior to surgical intervention. It is also a highly specific staging strategy that allows patients to be triaged to ALND based on a positive result (positive predictive value approximates 100%), thus avoiding two-stage axillary surgery and unnecessary SNB. Axillary UNB has a good clinical utility: based on an updated meta-analysis, we found that a median proportion of 18.4% (inter-quartile range=13.3%-27.4%) from 7,097 patients can be effectively triaged to axillary treatment and can avoid SNB. However, the changing algorithm of axillary surgical treatment means that UNB will have relatively less utility where surgeons omit ALND for minimal nodal metastatic disease. Research that allows enhanced application of ultrasound and UNB to specifically identify and biopsy sentinel nodes and to discriminate between patients with minimal versus advanced nodal metastatic involvement is likely to have the most impact on future management of the axilla in breast cancer.

Value of preoperative ultrasound-guided axillary lymph node biopsy for preventing completion axillary lymph node dissection in breast cancer: a systematic review and meta-analysis.

PURPOSE: This meta-analysis was designed to evaluate the utility of preoperative axillary ultrasound combined with US-guided lymph node biopsy if indicated (AUS ± biopsy), in terms of staging the axilla and preventing two-step axillary surgery in the form of sentinel node biopsy (SNB) followed by completion axillary lymph node (ALN) dissection. METHODS: We systematically searched electronic databases for studies that addressed preoperative assessment of ALN status by AUS ± biopsy. A pooled estimate was calculated for the false-negative rate (FNR) of AUS ± biopsy (defined as the proportion of women with a negative AUS ± biopsy result subsequently proven to have a positive axilla) and sensitivity (defined as the proportion of women with a positive AUS ± biopsy result among all women with a tumor positive axilla). RESULTS: The pooled FNR was 25 % (95 % confidence interval [CI] = 24-27) and the pooled sensitivity was 50 % (95 % CI = 43-57). There was substantial heterogeneity across studies for both FNR (I (2) = 69.42) and sensitivity (I (2) = 93.25), which was not explained by between-study differences in biopsy technique, mean/median tumor size, biopsy indication, or study design. Sensitivity was increased in studies with a high prevalence of ALN metastases. CONCLUSIONS: Preoperative axillary ultrasound-guided biopsy is a useful step in the process of axillary staging. Approximately 50 % of women with axillary involvement can be identified preoperatively. Still, one in four women with an ultrasound-guided biopsy-"proven" negative axilla has a positive SNB.

Development and evaluation of a prediction model for underestimated invasive breast cancer in women with ductal carcinoma in situ at stereotactic large core needle biopsy.

BACKGROUND: We aimed to develop a multivariable model for prediction of underestimated invasiveness in women with ductal carcinoma in situ at stereotactic large core needle biopsy, that can be used to select patients for sentinel node biopsy at primary surgery. METHODS: From the literature, we selected potential preoperative predictors of underestimated invasive breast cancer. Data of patients with nonpalpable breast lesions who were diagnosed with ductal carcinoma in situ at stereotactic large core needle biopsy, drawn from the prospective COBRA (Core Biopsy after RAdiological localization) and COBRA2000 cohort studies, were used to fit the multivariable model and assess its overall performance, discrimination, and calibration. RESULTS: 348 women with large core needle biopsy-proven ductal carcinoma in situ were available for analysis. In 100 (28.7%) patients invasive carcinoma was found at subsequent surgery. Nine predictors were included in the model. In the multivariable analysis, the predictors with the strongest association were lesion size (OR 1.12 per cm, 95% CI 0.98-1.28), number of cores retrieved at biopsy (OR per core 0.87, 95% CI 0.75-1.01), presence of lobular cancerization (OR 5.29, 95% CI 1.25-26.77), and microinvasion (OR 3.75, 95% CI 1.42-9.87). The overall performance of the multivariable model was poor with an explained variation of 9% (Nagelkerke's R(2)), mediocre discrimination with area under the receiver operating characteristic curve of 0.66 (95% confidence interval 0.58-0.73), and fairly good calibration. CONCLUSION: The evaluation of our multivariable prediction model in a large, clinically representative study population proves that routine clinical and pathological variables are not suitable to select patients with large core needle biopsy-proven ductal carcinoma in situ for sentinel node biopsy during primary surgery.

Investigation of lipid composition of dissected sentinel lymph nodes of breast cancer patients by 7T proton MR spectroscopy.

PURPOSE: To determine lipid composition of excised healthy and metastatic sentinel lymph nodes of breast cancer patients, as lipids are a potential discriminatory marker for malignancy. MATERIALS AND METHODS: Ten breast cancer patients undergoing surgical nodal staging were included. (1)H-magnetic resonance spectroscopic images (MRSI) were acquired without water and lipid suppression (resolution 3.0 × 3.0 × 5.0 mm(3)). MRSI was compared to histopathology. Six groups of lipid resonances (5.4-5.2, 4.3-4.1, 2.8, 2.3-2.0, 1.6-1.3, 0.9 ppm) were identified. The intensity ratios of the total of these resonances to this total including the water resonance and of each lipid resonance to the total of all lipid resonances were determined. For statistical analysis, a mixed model was applied after logistic transformation. The results were expressed as ratios of the median values of these lipid compositions in metastatic to benign nodes. RESULTS: In all, 6/32 (19%) of the excised nodes contained metastases. The ratios of the lipid resonances 5.4-5.2, 4.3-4.1, 2.8, 2.3-2.0, 1.6-1.3, 0.9 ppm between metastatic vs. benign were 0.3, 1.2, 0.2, 0.2, 1.2, and 0.9, respectively. Only the ratios of signals from unsaturated fatty acids to the total lipid signal differed significantly. CONCLUSION: Metastatic axillary lymph nodes contained fewer unsaturated fatty acids than benign nodes. 7T (1)H-MRS may be useful for detecting axillary breast cancer metastases.

Radiofrequency-assisted intact specimen biopsy of breast tumors: critical evaluation according to the IDEAL recommendations.

Radiofrequency-assisted intact specimen biopsy (RFIB) has been introduced for percutaneous biopsy or removal of breast tumors. Using radiofrequency cutting, the system enables the radiologist to obtain an intact sample of the target lesion. According to the IDEAL recommendations, we performed a critical evaluation of our initial experience with RFIB. Between June and November 2010, radiography-guided RFIB was performed in 19 female patients. All patients presented with suspicious microcalcifications (BI-RADS III-V) on mammography. Biopsy specimen integrity, thermal damage and histologic diagnosis were assessed by an expert breast pathologist. Data on technical success, diagnostic and therapeutic accuracy and periprocedural complications were collected and analyzed. The median age of the patients was 59 years. Median lesion diameter on mammography was 8 mm (range 2-76 mm). The procedure was successful in 16/19 (84%) patients and unsuccessful in 3/19 (16%) patients (2 non-representative samples, 1 sample with extensive thermal damage). Histologic analysis of the RFIB specimen revealed 12/19 (63%) benign lesions and 7/19 (37%) malignancies (4 ductal carcinoma in situ (DCIS) lesions and 3 invasive ductal carcinomas). In 1 patient, a DCIS lesion was completely removed with RFIB. Overall, 3 periprocedural complications occurred (1 wound leakage, 1 arterial hemorrhage and 1 infection requiring oral antibiotics). Tissue sampling of suspicious breast lesions can be performed successfully with RFIB. In 1 patient DCIS was radically excised with RFIB, which illustrates its potential as a minimally invasive therapeutic procedure for removal of small breast tumors. This is an interesting focus for further research when larger probe sizes become available.