Coagulation of human prostate volumes with MRI-controlled transurethral ultrasound therapy: results in gel phantoms.

PURPOSE: The feasibility and safety of magnetic resonance imaging (MRI)-controlled transurethral ultrasound therapy were demonstrated recently in a preliminary human study in which a small subvolume of prostate tissue was treated prior to radical prostatectomy. Translation of this technology to full clinical use, however, requires the capability to generate thermal coagulation in a volume up to that of the prostate gland itself. The aim of this study was to investigate the parameters required to treat a full 3D human prostate accurately with a multi-element transurethral applicator and multiplanar MR temperature control. METHODS: The approach was a combination of simulations (to select appropriate parameters) followed by experimental confirmation in tissue-mimicking phantoms. A ten-channel, MRI-compatible transurethral ultrasound therapy system was evaluated using six human prostate models (average volume: 36 cm(3)) obtained from the preliminary human feasibility study. Real-time multiplanar MR thermometry at 3 T was used to control the spatial heating pattern in up to nine planes simultaneously. Treatment strategies incorporated both single (4.6 or 8.1 MHz) and dual (4.6 and 14.4 MHz) frequencies, as well as maximum acoustic surface powers of 10 or 20 W cm(-2). RESULTS: Treatments at 4.6 MHz were capable of coagulating a volume equivalent to 97% of the prostate. Increasing power from 10 to 20 W cm(-2) reduced treatment times by approximately 50% with full treatments taking 26 ± 3 min at a coagulation rate of 1.8 ± 0.4 cm(3) min(-1). A dual-frequency 4.6∕14.4 MHz treatment strategy was shown to be the most effective configuration for achieving full human prostate treatment while maintaining good treatment accuracy for small treatment radii. The dual-frequency approach reduced overtreatment close to the prostate base and apex, confirming the simulations. CONCLUSIONS: This study reinforces the capability of MRI-controlled transurethral ultrasound therapy to treat full prostate volumes in a short treatment time with good spatial targeting accuracy and provides key parameters necessary for the next clinical trial.

Caffe con Troll: Shallow Ideas to Speed Up Deep Learning.

We present Caffe con Troll (CcT), a fully compatible end-to-end version of the popular framework Caffe with rebuilt internals. We built CcT to examine the performance characteristics of training and deploying general-purpose convolutional neural networks across different hardware architectures. We find that, by employing standard batching optimizations for CPU training, we achieve a 4.5× throughput improvement over Caffe on popular networks like CaffeNet. Moreover, with these improvements, the end-to-end training time for CNNs is directly proportional to the FLOPS delivered by the CPU, which enables us to efficiently train hybrid CPU-GPU systems for CNNs.