Nearest Neighbor-Based Strategy to Optimize Multi-View Triplet Network for Classification of Small-Sample Medical Imaging Data.

Multi-view classification with limited sample size and data augmentation is a very common machine learning (ML) problem in medicine. With limited data, a triplet network approach for two-stage representation learning has been proposed. However, effective training and verifying the features from the representation network for their suitability in subsequent classifiers are still unsolved problems. Although typical distance-based metrics for the training capture the overall class separability of the features, the performance according to these metrics does not always lead to an optimal classification. Consequently, an exhaustive tuning with all feature-classifier combinations is required to search for the best end result. To overcome this challenge, we developed a novel nearest-neighbor (NN) validation strategy based on the triplet metric. This strategy is supported by a theoretical foundation to provide the best selection of the features with a lower bound of the highest end performance. The proposed strategy is a transparent approach to identify whether to improve the features or the classifier. This avoids the need for repeated tuning. Our evaluations on real-world medical imaging tasks (i.e., radiation therapy delivery error prediction and sarcoma survival prediction) show that our strategy is superior to other common deep representation learning baselines [i.e., autoencoder (AE) and softmax]. The strategy addresses the issue of feature's interpretability which enables more holistic feature creation such that the medical experts can focus on specifying relevant data as opposed to tedious feature engineering.

Functional interrogation of adult hypothalamic neurogenesis with focal radiological inhibition.

The functional characterization of adult-born neurons remains a significant challenge. Approaches to inhibit adult neurogenesis via invasive viral delivery or transgenic animals have potential confounds that make interpretation of results from these studies difficult. New radiological tools are emerging, however, that allow one to noninvasively investigate the function of select groups of adult-born neurons through accurate and precise anatomical targeting in small animals. Focal ionizing radiation inhibits the birth and differentiation of new neurons, and allows targeting of specific neural progenitor regions. In order to illuminate the potential functional role that adult hypothalamic neurogenesis plays in the regulation of physiological processes, we developed a noninvasive focal irradiation technique to selectively inhibit the birth of adult-born neurons in the hypothalamic median eminence. We describe a method for Computer tomography-guided focal irradiation (CFIR) delivery to enable precise and accurate anatomical targeting in small animals. CFIR uses three-dimensional volumetric image guidance for localization and targeting of the radiation dose, minimizes radiation exposure to nontargeted brain regions, and allows for conformal dose distribution with sharp beam boundaries. This protocol allows one to ask questions regarding the function of adult-born neurons, but also opens areas to questions in areas of radiobiology, tumor biology, and immunology. These radiological tools will facilitate the translation of discoveries at the bench to the bedside.

Intraoperative fluoroscopic dose assessment in prostate brachytherapy patients.

PURPOSE: To evaluate a fluoroscopy-based intraoperative dosimetry system to guide placement of additional sources to underdosed areas, and perform computed tomography (CT) verification. METHODS AND MATERIALS: Twenty-six patients with prostate carcinoma treated with either I-125 or Pd-103 brachytherapy at the Puget Sound VA using intraoperative postimplant dosimetry were analyzed. Implants were performed by standard techniques. After completion of the initial planned brachytherapy procedure, the initial fluoroscopic intraoperative dose reconstruction analysis (I-FL) was performed with three fluoroscopic images acquired at 0 (AP), +15, and -15 degrees. Automatic seed identification was performed and the three-dimensional (3D) seed coordinates were computed and imported into VariSeed for dose visualization. Based on a 3D assessment of the isodose patterns additional seeds were implanted, and the final fluoroscopic intraoperative dose reconstruction was performed (FL). A postimplant computed tomography (CT) scan was obtained after the procedure and dosimetric parameters and isodose patterns were analyzed and compared. RESULTS: An average of 4.7 additional seeds were implanted after intraoperative analysis of the dose coverage (I-FL), and a median of 5 seeds. After implantation of additional seeds the mean V100 increased from 89% (I-FL) to 92% (FL) (p < 0.001). In I-125 patients an improvement from 91% to 94% (p = 0.01), and 87% to 93% (p = 0.001) was seen for Pd-103. The D90 increased from 105% (I-FL) to 122% (FL) (p < 0.001) for I-125, and 92% (I-FL) to 102% (FL) (p = 0.008) for Pd-103. A minimal change occurred in the R100 from a mean of 0.32 mL (I-FL) to 0.6 mL (FL) (p = 0.19). No statistical difference was noted in the R100 (rectal volume receiving 100% of the prescribed dose) between the two techniques. The rate of adverse isodose patterns decreased between I-FL and FL from 42% to 8%, respectively. The I-125 patients demonstrated a complete resolution of adverse isodose patterns after the initial isodose reconstruction (I-FL). The Pd-103 patients demonstrated a final rate of 8% gaps, 0% islands, and 0% holes on corrected isodose reconstruction. CONCLUSION: The use of intraoperative fluoroscopy-based dose assessment can accurately guide in the implantation of additional sources to supplement inadequately dosed areas within the prostate gland. Additionally, guided implantation of additional source, can significantly improve V100s and D90s, without significantly increasing rectal doses.