Deep Learning for Instance Retrieval: A Survey.

In recent years a vast amount of visual content has been generated and shared from many fields, such as social media platforms, medical imaging, and robotics. This abundance of content creation and sharing has introduced new challenges, particularly that of searching databases for similar content - Content Based Image Retrieval (CBIR) - a long-established research area in which improved efficiency and accuracy are needed for real-time retrieval. Artificial intelligence has made progress in CBIR and has significantly facilitated the process of instance search. In this survey we review recent instance retrieval works that are developed based on deep learning algorithms and techniques, with the survey organized by deep feature extraction, feature embedding and aggregation methods, and network fine-tuning strategies. Our survey considers a wide variety of recent methods, whereby we identify milestone work, reveal connections among various methods and present the commonly used benchmarks, evaluation results, common challenges, and propose promising future directions.

Monitoring of the repair process of surgically created lesions in equine superficial digital flexor tendons by use of computerized ultrasonography.

OBJECTIVE: To evaluate quantitative ultrasonography for objective monitoring of the healing process and prognostication of repair quality in equine superficial digital flexor (SDF) tendons. ANIMALS: 6 horses with standardized surgical lesions in SDF tendons of both forelimbs. PROCEDURES: Healing was monitored for 20 weeks after surgery by use of computerized ultrasonography. Pixels were categorized as C (intact fasciculi), B (incomplete fasciculi), E (accumulations of cells and fibrils), or N (homogenous fluid or cells). Four scars with the best quality of repair (repair group) and 4 scars with the lowest quality (inferior repair group) were identified histologically. Ratios for C, B, E, and N in both groups were compared. RESULTS: During 4 weeks after surgery, lesions increased 2- to 4-fold in length and 10-fold in volume. Until week 3 or 4, structure-related C and B ratios decreased sharply, whereas E and N ratios increased. After week 4, C and B ratios increased with gradually decreasing E and N ratios. At week 12, C and B ratios were equivalent. After week 12, C ratio increased slowly, but B ratio more rapidly. At week 20, C ratio remained constant, B ratio was substantially increased, and E and N ratios decreased. Values for the inferior repair group were most aberrant from normal. Ratios for C differed significantly between repair and inferior repair groups at weeks 16 and 18 and for B beginning at 14 weeks. CONCLUSIONS AND CLINICAL RELEVANCE: Computerized ultrasonography provided an excellent tool for objective monitoring of healing tendons in horses and reliable prognostication of repair quality.

Computerized ultrasonographic tissue characterization of equine superficial digital flexor tendons by means of stability quantification of echo patterns in contiguous transverse ultrasonographic images.

OBJECTIVE: To describe a method of computerized ultrasonographic tissue characterization that includes structures below the size limits of resolution in equine superficial digital flexor tendons. SAMPLE POPULATION: 2 damaged and 2 structurally normal superficial digital flexor tendons. PROCEDURE: Transverse ultrasonographic images were collected along the tendon long axis. Stability of echo pattern was quantified by means of variation in gray levels of each pixel in contiguous images and expressed as correlation, entropy, and waviness ratios. RESULTS: Normal young and normal old tissues were characterized by high correlation and low entropy and waviness ratios. In necrotic tissue, collapsed intratendinous septa resulted in high correlation, moderate entropy, and high waviness ratios. In early granulation tissue, complete lack of bundle formation resulted in values of zero for correlation and waviness ratios; loose connective tissue matrix resulted in a high entropy ratio. In late granulation tissue, formation of new bundles resulted in a high correlation ratio; swollen intratendinous septa and incomplete organization of connective tissue matrix were reflected in high entropy and waviness ratios. In early fibrotic tissue, rearrangement of tendon bundles resulted in a correlation ratio within reference range and a slight increase in the waviness ratio; an increase in cellularity and lack of fibrillar arrangement led to an increase in the entropy ratio. In late fibrotic and scar tissues, inferior quality of repair with almost complete lack of organization was reflected in low to moderate correlation, low waviness, and high entropy ratios. CONCLUSIONS AND CLINICAL RELEVANCE: Stability of echo patterns accurately reflects homogeneity of tendons in horses.