Exploring Generalizable Distillation for Efficient Medical Image Segmentation.

Efficient medical image segmentation aims to provide accurate pixel-wise predictions with a lightweight implementation framework. However, existing lightweight networks generally overlook the generalizability of the cross-domain medical segmentation tasks. In this paper, we propose Generalizable Knowledge Distillation (GKD), a novel framework for enhancing the performance of lightweight networks on cross-domain medical segmentation by generalizable knowledge distillation from powerful teacher networks. Considering the domain gaps between different medical datasets, we propose the Model-Specific Alignment Networks (MSAN) to obtain the domain-invariant representations. Meanwhile, a customized Alignment Consistency Training (ACT) strategy is designed to promote the MSAN training. Based on the domain-invariant vectors in MSAN, we propose two generalizable distillation schemes, Dual Contrastive Graph Distillation (DCGD) and Domain-Invariant Cross Distillation (DICD). In DCGD, two implicit contrastive graphs are designed to model the intra-coupling and inter-coupling semantic correlations. Then, in DICD, the domain-invariant semantic vectors are reconstructed from two networks (i.e., teacher and student) with a crossover manner to achieve simultaneous generalization of lightweight networks, hierarchically. Moreover, a metric named Fréchet Semantic Distance (FSD) is tailored to verify the effectiveness of the regularized domain-invariant features. Extensive experiments conducted on the Liver, Retinal Vessel and Colonoscopy segmentation datasets demonstrate the superiority of our method, in terms of performance and generalization ability on lightweight networks.

Micro-fabricated perforated polymer devices for long-term drug delivery.

Fabrication techniques have been developed to produce a perforated polymer microtube as a drug delivery device. The technique consists of first forming a silicon platform with trenches and alignment marks to hold the tubes for subsequent processing. Photolithography and reactive ion etching with an inductively coupled plasma source were used to fabricate micro holes on the surface of polyimide tubes. Several materials have been used to form the etching mask, including titanium film deposited by e-beam evaporation and SiO(2) and SiN(x) films deposited by high-density plasma chemical vapor deposition (HDPCVD). Three equidistant holes of 20 µm in diameter were fabricated on polyimide tubes (I.D. = 125 µm). The perforated tubes were loaded with ethinyl estradiol and tested for drug release in phosphate buffered saline (pH = 7.1) at 37°C. Zero order release was observed over a period of 30 days with a potential to be extended to 4 years.

Development and characterization of a scalable microperforated device capable of long-term zero order drug release.

A drug delivery system that consists of microperforated polyimide microtubes was developed and characterized. Two groups of polyimide tubes were used. One set consisted of microtubes (I.D. = 125 microm) with 32.9 +/- 1.7 microm size holes. The second set consisted of larger tubes (I.D. = 1000 microm) with 362-542 microm holes. The number of holes was varied between 1 and 3. The small tubes were loaded with crystal violet (CV) and ethinyl estradiol (EE) and the drug release studies were performed in 0.01 M phosphate buffered saline (PBS) (pH 7.1-7.4) at 37.0 +/- 1.0 degrees C for upto 4 weeks. The large tubes were loaded with CV and the drug release was studied in vitro in PBS and also ex vivo in rabbit's vitreous humor. Linear release rates with R(2) > 0.9900 were obtained for all groups with CV and EE. Release rates of 7.8 +/- 2.5, 16.2 +/- 5.5, and 22.5 +/- 6.0 ng/day for CV and 30.1 +/- 5.8 ng/day for EE were obtained for small tubes. For large tubes, a release rate of 10.8 +/- 4.1, 15.8 +/- 4.8 and 22.1 +/- 6.7 microg/day was observed in vitro in PBS and a release rate of 5.8 +/- 1.8 microg/day was observed ex vivo in vitreous humor.