PACS-integrated machine learning breast density classifier: clinical validation.

OBJECTIVE: To test the performance of a novel machine learning-based breast density tool. The tool utilizes a convolutional neural network to predict the BI-RADS based density assessment of a study. The clinical density assessments of 33,000 mammographic examinations (164,000 images) from one academic medical center (Site A) were used for training. MATERIALS AND METHODS: This was an IRB approved HIPAA compliant study performed at two academic medical centers. The validation data set was composed of 500 studies from one site (Site A) and 700 from another (Site B). At Site A, each study was assessed by three breast radiologists and the majority (consensus) assessment was used as truth. At Site B, if the tool agreed with the clinical reading, then it was considered to have correctly predicted the clinical reading. In cases where the tool and the clinical reading disagreed, then the study was evaluated by three radiologists and the consensus reading was used as the clinical reading. RESULTS: For the classification into the four categories of the Breast Imaging Reporting and Data System (BI-RADS®), the AI classifier had an accuracy of 84.6% at Site A and 89.7% at Site B. For binary classification (dense vs. non-dense), the AI classifier had an accuracy of 94.4% at Site A and 97.4% at Site B. In no case did the classifier disagree with the consensus reading by more than one category. CONCLUSIONS: The automated breast density tool showed high agreement with radiologists' assessments of breast density.

A practical approach to thermography in a hyperthermia/magnetic resonance hybrid system: validation in a heterogeneous phantom.

PURPOSE: This study investigates the feasibility and accuracy of noninvasive magnetic resonance (MR) monitoring for a system that includes a multiantenna applicator for part-body hyperthermia (SIGMA-Eye applicator, BSD-2000/3D) and a 1.5 Tesla MR tomograph (Siemens Magnetom Symphony). METHODS: A careful electrical decoupling enabled simultaneous operation of both systems, the hyperthermia system (100 MHz, up to 1600 W) and the MR tomograph (63.9 MHz). We used the phase data sets of a gradient echo sequence (long echo time TE = 20 ms) according to the proton frequency shift (PFS) method to determine MR temperature changes. Data postprocessing and visualization was conducted in the software platform AMIRA-HyperPlan. Heating was evaluated in an elliptical Lucite cylinder of 50 cm length filled with tissue-equivalent agarose and a skeleton made from low-dielectric material to simulate the heterogeneity of a real patient. Multiple catheters were included longitudinally for direct thermometry (using Bowman high-impedance thermistors). The phantom was positioned in the 24-antenna applicator SIGMA-Eye employing the integrated water bolus (filled with deionized water) both for coupling the radiated power into the lossy medium and to enable a correction procedure based on direct temperature measurements. RESULTS: In eight phantom experiments we monitored the heating in the applicator not only by repetitive acquisition of three-dimensional MR datasets, but also by measuring temperature-time curves directly at selected spatial positions. For the correction, we specified regions in the bolus. Direct bolus temperatures at fixed positions were taken to aim at best possible agreement between MR temperatures and these direct temperature-time curves. Then we compared additional direct temperature-position scans (thermal maps) for each experiment with the MR temperatures along these probes, which agreed satisfactorily (averaged accuracy of +/- 0.4-0.5 degrees C). The deviations decreased with decreasing observation time, temperature increase, and thermal load to the surroundings (corresponding to bolus heating)-estimating a resolution of, at best, +/- 0.2-0.3 degrees C. The acquired MR temperature distributions give also insight into limitations and control possibilities of regional hyperthermia (annular phased array technology) for various tumor sites. CONCLUSIONS: On-line MR monitoring of regional hyperthermia by using the PFS method is feasible in a phantom setup and can be further developed for clinical applications.

Visualization of the variability of 3D statistical shape models by animation.

Models of the 3D shape of anatomical objects and the knowledge about their statistical variability are of great benefit in many computer assisted medical applications like images analysis, therapy or surgery planning. Statistical model of shapes have successfully been applied to automate the task of image segmentation. The generation of 3D statistical shape models requires the identification of corresponding points on two shapes. This remains a difficult problem, especially for shapes of complicated topology. In order to interpret and validate variations encoded in a statistical shape model, visual inspection is of great importance. This work describes the generation and interpretation of statistical shape models of the liver and the pelvic bone.

Automatic segmentation of the liver for preoperative planning of resections.

This work presents first quantitative results of a method for automatic liver segmentation from CT data. It is based on a 3D deformable model approach using a-priori statistical information about the shape of the liver gained from a training set. The model is adapted to the data in an iterative process by analysis of the grey value profiles along its surface normals after nonlinear diffusion filtering. Leave-one-out experiments over 26 CT data sets reveal an accuracy of 2.4 mm with respect to the manual segmentation.

Estimation of heat transfer and temperature rise in partial-body regions during MR procedures: an analytical approach with respect to safety considerations.

In order to assess thermal response to RF exposure during MR procedures at the tissue level, simple analytical solutions to the non-stationary Pennes' bio-heat equation were obtained using the Green's function approach. Two thermal models appropriate for partial-body exposure were analyzed: In the first model, the temperature field at the periphery of an idealized volume RF resonator was modeled. The analytical solution reveals that tissue response to RF heating is characterized by an equilibration time and length. Both parameters are inversely related to tissue perfusion and vary for the soft-tissues considered between 0.27-25 min and 1.5-12 mm, respectively. None of the tissues investigated increase in temperature more than 0.5 degrees C for each W/kg of power dissipated. Secondly, a homogeneous tissue solution was derived that predicts the temperature-time course to an MR examination with time-varying specific absorption rates (SAR). Since SAR limits indicated in current MR safety standards relate to running SAR averages computed over an appropriate period of time, an expression was formulated that gives an upper limit for the temperature rise averaged over the same period of time, as a function of both the upper limit of running SAR averages and the duration of the MR examination. The analysis revealed that the partial-body SAR limits indicated in the IEC standard may not guarantee under all circumstances compliance with the basic restrictions concerning temperature rise.