Mammography Breast Cancer Screening Triage Using Deep Learning: A UK Retrospective Study.

Background Breast screening enables early detection of cancers; however, most women have normal mammograms, resulting in repetitive and resource-intensive reading tasks. Purpose To investigate if deep learning (DL) algorithms can be used to triage mammograms by identifying normal results to reduce workload or flag cancers that may be overlooked. Materials and Methods In this retrospective study, three commercial DL algorithms were investigated using consecutive mammograms from two UK Breast Screening Program sites from January 2015 to December 2017 and January 2017 to December 2018 on devices from two mammography vendors. Normal mammograms with a 3-year follow-up and histopathologically proven cancer detected at screening, the subsequent round, or in the 3-year interval were included. Two algorithm thresholds were set: in scenario A, 99.0% sensitivity for rule-out triage to a lone reader, and in scenario B, approximately 1.0% additional recall providing a rule-in triage for further assessment. Both thresholds were then applied to the screening workflow in scenario C. The sensitivity and specificity were used to assess the overall predictive performance of each DL algorithm. Results The data set comprised 78 849 patients (median age, 59 years [IQR, 53-63 years]) and 887 screening-detected, 439 interval, and 688 subsequent screening round-detected cancers. In scenario A (rule-out triage), models DL-1, DL-2, and DL-3 triaged 35.0% (27 565 of 78 849), 53.2% (41 937 of 78 849), and 55.6% (43 869 of 78 849) of mammograms, respectively, with 0.0% (0 of 887) to 0.1% (one of 887) of screening-detected cancers undetected. In scenario B, DL algorithms triaged in 4.6% (20 of 439) to 8.2% (36 of 439) of interval and 5.2% (36 of 688) to 6.1% (42 of 688) of subsequent-round cancers when applied after the routine double-reading workflow. Combining both approaches in scenario C resulted in an overall noninferior specificity (difference, -0.9%; P < .001) and superior sensitivity (difference, 2.7%; P < .001) for the adaptive workflow compared with routine double reading for all three algorithms. Conclusion Rule-out and rule-in DL-adapted triage workflows can improve the efficiency and efficacy of mammography breast cancer screening. © RSNA, 2023 Supplemental material is available for this article. See also the editorial by Nishikawa and Lu in this issue.

Improved outcome in penile cancer with radiologically enhanced stratification protocol for lymph node staging procedures: a study in 316 inguinal basins with a mean follow-up of 5 years.

BACKGROUND: Lymph node metastasis is the main determinant of survival in penile cancer patients. Conventionally clinical palpability is used to stratify patients to Inguinal Lymph node dissection (ILND) if clinically node positive (cN +) or Dynamic sentinel node biopsy (DSNB) if clinically node negative (cN0). Studies suggest a false negative rate (FNR) of around 10% (5-13%) for DSNB. To our knowledge there are no studies reporting harder end point of survival and outcomes of all clinically node positive (cN +) patients. We present our outcome data of all patients with penile cancer including false negative rates and survival in both DSNB and ILND groups. METHODS: One hundred fifty-eight consecutive patients (316 inguinal basins), who had lymph node surgery for penile cancer in a tertiary referral centre from Jan 2008 to 2018, were included in the study. All patients underwent ultrasound (US) ± fine needle aspiration cytology (FNAC) and then MRI/ CT, if needed, to stage their disease. We used combined clinical and radiological criteria (node size, architecture loss, irregular margins) to stratify patients to DSNB vs ILND as opposed to clinical palpability alone. RESULTS: 11.2% i.e., 27/241 inguinal basins had lymph node positive disease by DSNB. 54.9% i.e., 39/71 inguinal basins (IBs) had lymph node-positive disease by ILND. 4 inguinal basins with no tracer uptake in sentinel node scans are being monitored at patient's request and have not had any recurrences to date. With a mean follow-up of 65 months (range 24-150), the false-negative rate (FNR) for DSNB is 0%. Judicious uses of cross-sectional imaging necessitated ILND in 2 inguinal basins with non-palpable nodes and negative US with false positive rate of 6.3% (2/32) for ILND. The same cohort of DSNB patients might have had 11.1% (3/27) FNR if only palpability criteria was used. 43 (28%) patients who did require cross sectional imaging as per our criteria had a low node positive rate of 4.7% (p = 0.03). Mean cancer specific survival of all node-positive patients was 105 months. CONCLUSION: The performance of DSNB improved with enhanced radiological stratification of patients to either DSNB or ILND. We for the first time report the comprehensive outcome of all lymph node staging procedures in penile cancer.

Radio-frequency identification (RFID) tag localisation of non-palpable breast lesions a single centre experience.

AIM: The purpose of this study is to report the surgical experience and outcomes with pre-operative localisation of non-palpable breast lesions using the RFID tag system. METHODS: The cohort for this prospective study included patients over the age of 18 with biopsy proven, non-palpable indeterminate lesions, DCIS or breast cancer requiring pre-operative localisation before surgical excision between September 2020 and July 2022. RESULTS: A total of 312 RFID tags were placed in 299 consecutive patients. Indications for localisation included non-palpable invasive cancer in 255 (85.3%) patients, in situ disease in 38 (12.7%) and indeterminate lesions requiring surgical excision in 6 (2.0%). Both in situ and invasive lesions had a median size of 13 mm (range 4-100 mm) on pre-operative imaging. The RFID tags were in situ for a median time of 21 days before surgery (range 0-233 days). Of the 213 tags, 292 (93.6%) were introduced using ultrasound (USS) guidance and stereotactically in 20 (6.4%). In 3 (1.0%) cases the RFID tag was either not satisfactorily deployed at the intended target or retrieved intra-operatively. Following discussion of post-operative histology by the multi-disciplinary team, further surgery for close or involved margins was for 26 (8.7%) patients. CONCLUSION: The Hologic RFID tag system can be used for accurate pre-operative localisation of non-palpable masses as well as diffuse abnormalities such as mammographic distortions and calcifications. It has advantages of flexibility for scheduling image-guided insertion independently of scheduled operating lists and can be placed to localise lesions prior to initiating neoadjuvant systemic treatment.

Radio-Frequency Identifier Devices (RFIDs): Our Experience With Wireless Localisation in Non-palpable Breast Masses at a UK Tertiary Breast Imaging Unit.

The aim of this report is to evaluate the impact of the percutaneous ultrasound-guided placement of wireless radio-frequency identifier devices (RFIDs; Hologic LOCalizer, Marlborough, Massachusett) and its impact in our practice of preoperative localisation of biopsy-proven breast cancers, post-vacuum assisted biopsy-site hematoma, and lymph nodes for targeted dissection pre-operatively. A single institutional retrospective analysis of RFID usage for preoperative localisation in screening and symptomatic patients with non-palpable biopsy-proven breast carcinoma was reviewed from the radiology information system at our tertiary breast imaging unit. Its impact on the radiological and surgical team practice was reviewed, including the number of appointments, the interval between scheduling image-guided localisation and intraoperative localisation, procedure failure, average deployment, and surgical time. Feedback from surgeons and pathologists practice was also taken into consideration. Fifty-nine RFID clips were placed for wireless localisation of breast cancers, lymph nodes, and post-vacuum-assisted biopsy hematoma over nine months. Seventy-three per cent (73%; n=43/59) of RFID devices were placed in biopsy-proven carcinomas under direct ultrasound guidance. The learning curve was small, as the delivery system was similar to the commonly used localisation clips. The pilot process involved RFID with radioisotope injections for the breast mass for the initial 28% (n=12 /43) cases, which were gradually transitioned into RFID only. Radioisotope was used for sentinel node purposes if required. For targeted node dissection, 3% (n=2/59) patients received RFID for biopsy-proven metastatic node localisation, with one of two with adjunct radioisotope injection. Post-vacuum-assisted biopsy (VAB) hematoma was localised in 24% (n=14/59) cases, four of which received adjunct radioisotope in the pilot phase. The average procedure time for RFID deployment was five minutes. The average time for surgery was 20 minutes. 1.6% (n=1/59) incidence of RFID slipping from the surface of the site through surgical exposure attributed to the superficial and immediate pre-operative placement of the RFID. This was salvageable with adjunct radioisotope injection within the pilot phase. There were no incidences of repeat localisation or repeat exploration surgeries. Planned pre-operative localisation with RFID allows for better planning and less pressured service delivery and a success rate of 98-99%. This ultimately avoids lost theatre time and patient demotivation. Surgeons have reported excellent intra-operative detectability in their approach. There has been no difficulty in the detection of the RFID within the surgical cavity despite hematoma. RFID localisers are expensive compared to our usual practice of radioisotope injection but this can be recuperated through uncoupled tariffs like gain in slots of one-stop clinics, flexibility for placement, and avoiding lost theatre time, as these can be placed up to 30 days before surgery.

Breast ultrasound region of interest detection and lesion localisation.

In current breast ultrasound computer aided diagnosis systems, the radiologist preselects a region of interest (ROI) as an input for computerised breast ultrasound image analysis. This task is time consuming and there is inconsistency among human experts. Researchers attempting to automate the process of obtaining the ROIs have been relying on image processing and conventional machine learning methods. We propose the use of a deep learning method for breast ultrasound ROI detection and lesion localisation. We use the most accurate object detection deep learning framework - Faster-RCNN with Inception-ResNet-v2 - as our deep learning network. Due to the lack of datasets, we use transfer learning and propose a new 3-channel artificial RGB method to improve the overall performance. We evaluate and compare the performance of our proposed methods on two datasets (namely, Dataset A and Dataset B), i.e. within individual datasets and composite dataset. We report the lesion detection results with two types of analysis: (1) detected point (centre of the segmented region or the detected bounding box) and (2) Intersection over Union (IoU). Our results demonstrate that the proposed methods achieved comparable results on detected point but with notable improvement on IoU. In addition, our proposed 3-channel artificial RGB method improves the recall of Dataset A. Finally, we outline some future directions for the research.

Breast ultrasound lesions recognition: end-to-end deep learning approaches.

Multistage processing of automated breast ultrasound lesions recognition is dependent on the performance of prior stages. To improve the current state of the art, we propose the use of end-to-end deep learning approaches using fully convolutional networks (FCNs), namely FCN-AlexNet, FCN-32s, FCN-16s, and FCN-8s for semantic segmentation of breast lesions. We use pretrained models based on ImageNet and transfer learning to overcome the issue of data deficiency. We evaluate our results on two datasets, which consist of a total of 113 malignant and 356 benign lesions. To assess the performance, we conduct fivefold cross validation using the following split: 70% for training data, 10% for validation data, and 20% testing data. The results showed that our proposed method performed better on benign lesions, with a top "mean Dice" score of 0.7626 with FCN-16s, when compared with the malignant lesions with a top mean Dice score of 0.5484 with FCN-8s. When considering the number of images with Dice score > 0.5 , 89.6% of the benign lesions were successfully segmented and correctly recognised, whereas 60.6% of the malignant lesions were successfully segmented and correctly recognized. We conclude the paper by addressing the future challenges of the work.

Correction to: Phase 2 Open-Label Trial Investigating Percutaneous Laser Ablation for Treatment of Early-Stage Breast Cancer: MRI, Pathology, and Outcome Correlations.

The article "Phase 2 Open-Label Trial Investigating Percutaneous Laser Ablation for Treatment of Early-Stage Breast Cancer: MRI, Pathology, and Outcome Correlations", written by Barbara Schwartzberg et al., was originally published electronically on the publisher's internet portal (currently SpringerLink) on July 9, 2018, without open access.

Phase 2 Open-Label Trial Investigating Percutaneous Laser Ablation for Treatment of Early-Stage Breast Cancer: MRI, Pathology, and Outcome Correlations.

BACKGROUND: An institutional review board-approved, multicenter clinical trial was designed to determine the efficacy and outcome of percutaneous laser ablation (PLA) in the treatment of invasive ductal breast carcinoma (IDC). Post-ablation magnetic resonance imaging (MRI) was compared with surgical pathology in evaluation of residual post-ablation IDC and ductal carcinoma in situ. METHODS: Patients with a single focus of IDC 20 mm or smaller by pre-ablation MRI were treated with PLA. The patients underwent a 28-day post-ablation MRI, followed by surgical resection. Cell viability criteria were applied to pre- and post-ablation pathology specimens, which evaluated hematoxylin-eosin (H&E), cytokeratin (CK) 8/18, estrogen receptor, and Ki67 staining patterns. RESULTS: In this study, 61 patients were reported as the intention-to-treat cohort for determination of PLA efficacy. Of these 61 patients, 51 (84%) had complete tumor ablation confirmed by pathology analysis. One subject's MRI imaging was not performed per protocol, which left 60 subjects evaluable for MRI pathology correlation. Five patients (8.3%) had residual IDC shown by both MRI and pathology. Post-ablation discordance was noted between MRI and pathology, with four patients (6.7%) false-positive and four patients (6.7%) false-negative. The negative predictive value (NPV) of MRI for all the patients was 92.2% (95% confidence interval [CI], 71.9-91.9%). Of the 47 patients (97.9%) with tumors 15 mm or smaller, 46 were completely ablated, with an MRI NPV of 97.7% (95% CI, 86.2-99.9%). CONCLUSIONS: Percutaneous laser ablation is a potential alternative to surgery for treatment of early-stage IDC. Strong correlations exist between post-ablation MRI and pathologic alterations in CK8/18, ER, and Ki67 staining.

Breast image pre-processing for mammographic tissue segmentation.

During mammographic image acquisition, a compression paddle is used to even the breast thickness in order to obtain optimal image quality. Clinical observation has indicated that some mammograms may exhibit abrupt intensity change and low visibility of tissue structures in the breast peripheral areas. Such appearance discrepancies can affect image interpretation and may not be desirable for computer aided mammography, leading to incorrect diagnosis and/or detection which can have a negative impact on sensitivity and specificity of screening mammography. This paper describes a novel mammographic image pre-processing method to improve image quality for analysis. An image selection process is incorporated to better target problematic images. The processed images show improved mammographic appearances not only in the breast periphery but also across the mammograms. Mammographic segmentation and risk/density classification were performed to facilitate a quantitative and qualitative evaluation. When using the processed images, the results indicated more anatomically correct segmentation in tissue specific areas, and subsequently better classification accuracies were achieved. Visual assessments were conducted in a clinical environment to determine the quality of the processed images and the resultant segmentation. The developed method has shown promising results. It is expected to be useful in early breast cancer detection, risk-stratified screening, and aiding radiologists in the process of decision making prior to surgery and/or treatment.

A Review on Automatic Mammographic Density and Parenchymal Segmentation.

Breast cancer is the most frequently diagnosed cancer in women. However, the exact cause(s) of breast cancer still remains unknown. Early detection, precise identification of women at risk, and application of appropriate disease prevention measures are by far the most effective way to tackle breast cancer. There are more than 70 common genetic susceptibility factors included in the current non-image-based risk prediction models (e.g., the Gail and the Tyrer-Cuzick models). Image-based risk factors, such as mammographic densities and parenchymal patterns, have been established as biomarkers but have not been fully incorporated in the risk prediction models used for risk stratification in screening and/or measuring responsiveness to preventive approaches. Within computer aided mammography, automatic mammographic tissue segmentation methods have been developed for estimation of breast tissue composition to facilitate mammographic risk assessment. This paper presents a comprehensive review of automatic mammographic tissue segmentation methodologies developed over the past two decades and the evidence for risk assessment/density classification using segmentation. The aim of this review is to analyse how engineering advances have progressed and the impact automatic mammographic tissue segmentation has in a clinical environment, as well as to understand the current research gaps with respect to the incorporation of image-based risk factors in non-image-based risk prediction models.

Lumbar lordosis and pars interarticularis fractures: a case-control study.

OBJECTIVE: The aim of this study is to examine the relationship between lumbar lordosis and pars interarticularis fractures. MATERIALS AND METHODS: In this retrospective case-control study we compare the angle of lumbar lordosis and the angle of the S1 vertebral endplate (as a measure of pelvic tilt) in patients with bilateral L5 pars interarticularis fractures with age- and sex-matched control cases with normal MRI examinations of the lumbar spine. Twenty-nine cases of bilateral L5 pars interarticularis fractures with matched control-cases were identified on MRI (16 male, 13 female, age 9-63 years). The angle of lordosis was measured between the inferior L4 and superior S1 vertebral endplates on a standing lateral lumbar spine radiograph for both groups. RESULTS: The mean angle of lordosis about the L5 vertebra was 36.9° (SD = 6.5°) in the pars interarticularis fracture group, and 30.1° (SD = 6.4°) in the control group. The difference between the two groups was significant (mean difference 6.8°, Student's t test: P < 0.001). The mean angle of sacral tilt measured was 122.2° (SD = 10.16°) for controls and 136.4° (SD = 10.86°) for patients with pars defects. The difference in the means of 14.2° was statistically significantly different (P < 0.0001). CONCLUSION: Sacral tilt represented by a steeply angled superior endplate of S1 is associated with a significantly increased angle of lordosis, between L4 and S1, and pars fractures at L5. Steep angulation of the first sacral vertebral segment maybe the predisposing biomechanical factor that leads to pincer-like impingement of the pars interarticularis and then spondylolysis.