Hydra: Multi-head low-rank adaptation for parameter efficient fine-tuning.

The recent surge in large-scale foundation models has spurred the development of efficient methods for adapting these models to various downstream tasks. Low-rank adaptation methods, such as LoRA, have gained significant attention due to their outstanding parameter efficiency and no additional inference latency. This paper investigates a more general form of adapter module based on the analysis that parallel and sequential adaptation branches learn novel and general features during fine-tuning, respectively. The proposed method, named Hydra, combines parallel and sequential branch to integrate capabilities, which is more expressive than existing single branch methods and enables the exploration of a broader range of optimal points in the fine-tuning process. In addition, the proposed method explicitly leverages the pre-trained weights by performing a linear combination of the pre-trained features. It allows the learned features to have better generalization performance across diverse downstream tasks. Furthermore, we perform a comprehensive analysis of the characteristics of each adaptation branch with empirical evidence. Through an extensive range of experiments, we substantiate the efficiency and demonstrate the superior performance of Hydra. This comprehensive evaluation underscores the potential impact and effectiveness of Hydra in a variety of applications. The source code of this work is publicly opened on https://github.com/extremebird/Hydra.

Digital histological staining of tissue slide images from optical coherence microscopy.

The convergence of staining-free optical imaging and digital staining technologies has become a central focus in digital pathology, presenting significant advantages in streamlining specimen preparation and expediting the rapid acquisition of histopathological information. Despite the inherent merits of optical coherence microscopy (OCM) as a staining-free technique, its widespread application in observing histopathological slides has been constrained. This study introduces a novel approach by combining wide-field OCM with digital staining technology for the imaging of histopathological slides. Through the optimization of the histology slide production process satisfying the ground growth for digital staining as well as pronounced contrast for OCM imaging, successful imaging of various mouse tissues was achieved. Comparative analyses with conventional staining-based bright field images were executed to evaluate the proposed methodology's efficacy. Moreover, the study investigates the generalization of digital staining color appearance to ensure consistent histopathology, considering tissue-specific and thickness-dependent variations.

Development of a tetrahydroindazolone-based HDAC6 inhibitor with in-vivo anti-arthritic activity.

Histone deacetylase 6 (HDAC6) induces the expression of pro-inflammatory cytokines in macrophages; therefore, HDAC inhibitors may be beneficial for the treatment of macrophage-associated immune disorders and chronic inflammatory diseases, including atherosclerosis and rheumatoid arthritis. Structure-activity relationship studies were conducted on various phenyl hydroxamate HDAC6 inhibitors with indolone/indazolone-based bi- or tricyclic ring moieties as the cap group aiming to develop novel anti-arthritic drug candidates. Several compounds exhibited nanomolar activity and HDAC6 selectivity greater than 500-fold over HDAC1. Compound 21, a derivative with the tetrahydroindazolone cap group, is a potent HDAC6 inhibitor with an IC50 of 18 nM and 217-fold selectivity over HDAC1 and showed favorable oral bioavailability in animals. Compound 21 increases the acetylation level of tubulin without affecting histone acetylation in cutaneous T-cell lymphoma cells and inhibits TNF-α secretion in LPS-stimulated macrophage cells. The anti-arthritic effects of compound 21 were evaluated using a rat adjuvant-induced arthritis (AIA) model. Treatment with compound 21 significantly reduced the arthritis score, and combination treatment with methotrexate showed a synergistic effect in AIA models. We identified a novel HDAC6 inhibitor, compound 21, with excellent in vivo anti-arthritic efficacy, which can lead to the development of oral anti-arthritic drugs.

Multi-contrast digital histopathology of mouse organs using quantitative phase imaging and virtual staining.

Quantitative phase imaging (QPI) has emerged as a new digital histopathologic tool as it provides structural information of conventional slide without staining process. It is also capable of imaging biological tissue sections with sub-nanometer sensitivity and classifying them using light scattering properties. Here we extend its capability further by using optical scattering properties as imaging contrast in a wide-field QPI. In our first step towards validation, QPI images of 10 major organs of a wild-type mouse have been obtained followed by H&E-stained images of the corresponding tissue sections. Furthermore, we utilized deep learning model based on generative adversarial network (GAN) architecture for virtual staining of phase delay images to a H&E-equivalent brightfield (BF) image analogues. Using the structural similarity index, we demonstrate similarities between virtually stained and H&E histology images. Whereas the scattering-based maps look rather similar to QPI phase maps in the kidney, the brain images show significant improvement over QPI with clear demarcation of features across all regions. Since our technology provides not only structural information but also unique optical property maps, it could potentially become a fast and contrast-enriched histopathology technique.

Deep Learning-Based Glaucoma Screening Using Regional RNFL Thickness in Fundus Photography.

Since glaucoma is a progressive and irreversible optic neuropathy, accurate screening and/or early diagnosis is critical in preventing permanent vision loss. Recently, optical coherence tomography (OCT) has become an accurate diagnostic tool to observe and extract the thickness of the retinal nerve fiber layer (RNFL), which closely reflects the nerve damage caused by glaucoma. However, OCT is less accessible than fundus photography due to higher cost and expertise required for operation. Though widely used, fundus photography is effective for early glaucoma detection only when used by experts with extensive training. Here, we introduce a deep learning-based approach to predict the RNFL thickness around optic disc regions in fundus photography for glaucoma screening. The proposed deep learning model is based on a convolutional neural network (CNN) and utilizes images taken with fundus photography and with RNFL thickness measured with OCT for model training and validation. Using a dataset acquired from normal tension glaucoma (NTG) patients, the trained model can estimate RNFL thicknesses in 12 optic disc regions from fundus photos. Using intuitive thickness labels to identify localized damage of the optic nerve head and then estimating regional RNFL thicknesses from fundus images, we determine that screening for glaucoma could achieve 92% sensitivity and 86.9% specificity. Receiver operating characteristic (ROC) analysis results for specificity of 80% demonstrate that use of the localized mean over superior and inferior regions reaches 90.7% sensitivity, whereas 71.2% sensitivity is reached using the global RNFL thicknesses for specificity at 80%. This demonstrates that the new approach of using regional RNFL thicknesses in fundus images holds good promise as a potential screening technique for early stage of glaucoma.

Encoder-Weighted W-Net for Unsupervised Segmentation of Cervix Region in Colposcopy Images.

Cervical cancer can be prevented and treated better if it is diagnosed early. Colposcopy, a way of clinically looking at the cervix region, is an efficient method for cervical cancer screening and its early detection. The cervix region segmentation significantly affects the performance of computer-aided diagnostics using a colposcopy, particularly cervical intraepithelial neoplasia (CIN) classification. However, there are few studies of cervix segmentation in colposcopy, and no studies of fully unsupervised cervix region detection without image pre- and post-processing. In this study, we propose a deep learning-based unsupervised method to identify cervix regions without pre- and post-processing. A new loss function and a novel scheduling scheme for the baseline W-Net are proposed for fully unsupervised cervix region segmentation in colposcopy. The experimental results showed that the proposed method achieved the best performance in the cervix segmentation with a Dice coefficient of 0.71 with less computational cost. The proposed method produced cervix segmentation masks with more reduction in outliers and can be applied before CIN detection or other diagnoses to improve diagnostic performance. Our results demonstrate that the proposed method not only assists medical specialists in diagnosis in practical situations but also shows the potential of an unsupervised segmentation approach in colposcopy.

Deep-Learning-Based Algorithm for the Removal of Electromagnetic Interference Noise in Photoacoustic Endoscopic Image Processing.

Despite all the expectations for photoacoustic endoscopy (PAE), there are still several technical issues that must be resolved before the technique can be successfully translated into clinics. Among these, electromagnetic interference (EMI) noise, in addition to the limited signal-to-noise ratio (SNR), have hindered the rapid development of related technologies. Unlike endoscopic ultrasound, in which the SNR can be increased by simply applying a higher pulsing voltage, there is a fundamental limitation in leveraging the SNR of PAE signals because they are mostly determined by the optical pulse energy applied, which must be within the safety limits. Moreover, a typical PAE hardware situation requires a wide separation between the ultrasonic sensor and the amplifier, meaning that it is not easy to build an ideal PAE system that would be unaffected by EMI noise. With the intention of expediting the progress of related research, in this study, we investigated the feasibility of deep-learning-based EMI noise removal involved in PAE image processing. In particular, we selected four fully convolutional neural network architectures, U-Net, Segnet, FCN-16s, and FCN-8s, and observed that a modified U-Net architecture outperformed the other architectures in the EMI noise removal. Classical filter methods were also compared to confirm the superiority of the deep-learning-based approach. Still, it was by the U-Net architecture that we were able to successfully produce a denoised 3D vasculature map that could even depict the mesh-like capillary networks distributed in the wall of a rat colorectum. As the development of a low-cost laser diode or LED-based photoacoustic tomography (PAT) system is now emerging as one of the important topics in PAT, we expect that the presented AI strategy for the removal of EMI noise could be broadly applicable to many areas of PAT, in which the ability to apply a hardware-based prevention method is limited and thus EMI noise appears more prominently due to poor SNR.

Polyphyllin D Shows Anticancer Effect through a Selective Inhibition of Src Homology Region 2-Containing Protein Tyrosine Phosphatase-2 (SHP2).

Natural products have continued to offer tremendous opportunities for drug development, as they have long been used in traditional medicinal systems. SHP2 has served as an anticancer target. To identify novel SHP2 inhibitors with potential anticancer activity, we screened a library containing 658 natural products. Polyphyllin D was found to selectively inhibit SHP2 over SHP1, whereas two other identified compounds (echinocystic acid and oleanolic acid) demonstrated dual SHP1 and SHP2 inhibition. In a cell-based assay, polyphyllin D exhibited cytotoxicity in Jurkat cells, an acute lymphoma leukemia cell line, whereas the other two compounds were ineffective. Polyphyllin D also decreased the level of phosphorylated extracellular signal-regulated kinase (p-ERK), a proliferation marker in Jurkat cells. Furthermore, knockdown of protein tyrosine phosphatase (PTP)N6 (SHP1) or PTPN11 (SHP2) decreased p-ERK levels. However, concurrent knockdown of PTPN6 and PTPN11 in Jurkat cells recovered p-ERK levels. These results demonstrated that polyphyllin D has potential anticancer activity, which can be attributed to its selective inhibition of SHP2 over SHP1.

Quantitative Screening of Cervical Cancers for Low-Resource Settings: Pilot Study of Smartphone-Based Endoscopic Visual Inspection After Acetic Acid Using Machine Learning Techniques.

BACKGROUND: Approximately 90% of global cervical cancer (CC) is mostly found in low- and middle-income countries. In most cases, CC can be detected early through routine screening programs, including a cytology-based test. However, it is logistically difficult to offer this program in low-resource settings due to limited resources and infrastructure, and few trained experts. A visual inspection following the application of acetic acid (VIA) has been widely promoted and is routinely recommended as a viable form of CC screening in resource-constrained countries. Digital images of the cervix have been acquired during VIA procedure with better quality assurance and visualization, leading to higher diagnostic accuracy and reduction of the variability of detection rate. However, a colposcope is bulky, expensive, electricity-dependent, and needs routine maintenance, and to confirm the grade of abnormality through its images, a specialist must be present. Recently, smartphone-based imaging systems have made a significant impact on the practice of medicine by offering a cost-effective, rapid, and noninvasive method of evaluation. Furthermore, computer-aided analyses, including image processing-based methods and machine learning techniques, have also shown great potential for a high impact on medicinal evaluations. OBJECTIVE: In this study, we demonstrate a new quantitative CC screening technique and implement a machine learning algorithm for smartphone-based endoscopic VIA. We also evaluated the diagnostic performance and practicability of the approach based on the results compared to the gold standard and from physicians' interpretation. METHODS: A smartphone-based endoscope system was developed and applied to the VIA screening. A total of 20 patients were recruited for this study to evaluate the system. Overall, five were healthy, and 15 were patients who had shown a low to high grade of cervical intraepithelial neoplasia (CIN) from both colposcopy and cytology tests. Endoscopic VIA images were obtained before a loop electrosurgical excision procedure for patients with abnormal tissues, and their histology tissues were collected. Endoscopic VIA images were assessed by four expert physicians relative to the gold standard of histopathology. Also, VIA features were extracted from multiple steps of image processing techniques to find the differences between abnormal (CIN2+) and normal (≤CIN1). By using the extracted features, the performance of different machine learning classifiers, such as k-nearest neighbors (KNN), support vector machine, and decision tree (DT), were compared to find the best algorithm for VIA. After determining the best performing classifying model, it was used to evaluate the screening performance of VIA. RESULTS: An average accuracy of 78%, with a Cohen kappa of 0.571, was observed for the evaluation of the system by four physicians. Through image processing, 240 sliced images were obtained from the cervicogram at each clock position, and five features of VIA were extracted. Among the three models, KNN showed the best performance for finding VIA within holdout 10-fold cross-validation, with an accuracy of 78.3%, area under the curve of 0.807, a specificity of 80.3%, and a sensitivity of 75.0%, respectively. The trained model performed using an unprovided data set resulted in an accuracy of 80.8%, specificity of 84.1%, and sensitivity of 71.9%. Predictions were visualized with intuitive color labels, indicating the normal/abnormal tissue using a circular clock-type segmentation. Calculating the overlapped abnormal tissues between the gold standard and predicted value, the KNN model overperformed the average assessments of physicians for finding VIA. CONCLUSIONS: We explored the potential of the smartphone-based endoscopic VIA as an evaluation technique and used the cervicogram to evaluate normal/abnormal tissue using machine learning techniques. The results of this study demonstrate its potential as a screening tool in low-resource settings.

Identification of CKD-516: a potent tubulin polymerization inhibitor with marked antitumor activity against murine and human solid tumors.

Tubulin polymerization inhibitors had emerged as one of promising anticancer therapeutics because of their dual mechanism of action, i.e. apoptosis by cell-cycle arrest and VDA, vascular disrupting agent. VDAs are believed to be more efficient, less toxic, and several of them are currently undergoing clinical trials. To identify novel tubulin inhibitors that possess potent cytotoxicity and strong inhibition of tubulin polymerization as well as potent in vivo antitumor efficacy, we have utilized benzophenone scaffold. Complete SAR analysis of newly synthesized analogues that were prepared by incorporation of small heterocycles (C2, C4, and C5 position) into B-ring along with the evaluation of their in vitro cytotoxicity, tubulin polymerization inhibition, and in vivo antitumor activity allowed us to identify 22 (S516). Compound 22 was found to have potent cytotoxicity against several cancer cells including P-gp overexpressing MDR positive cell line (HCT15). It also induced cell cycle arrest at G(2)/M phase, which is associated with strong inhibition of tubulin polymerization. Its in vivo efficacy was improved by preparing its (l)-valine prodrug, 65 (CKD-516), which together with greatly improved aqueous solubility has shown marked antitumor efficacy against both murine tumors (CT26 and 3LL) and human xenogratfs (HCT116 and HCT15) in mice.

The scope of thallium nitrate oxidative cyclization of chalcones; synthesis and evaluation of isoflavone and aurone analogs for their inhibitory activity against interleukin-5.

The oxidative cyclization of 2'-hydroxy-6'-cyclohexylmethoxychalcones 5 using thallium (III) nitrate (TTN) in alcoholic solvents produced isoflavones 2 and (or) aurones 3 depending on the electronic nature of p-substituents on ring B. Chalcones with strong electron donating substituents (OH, OCH(3)) were exclusively converted to isoflavones 2. Chalcone with weak electron donating substituents (CH(2)CH(3)) was transformed into isoflavone 2 and the aurone 3 in approximate ratio 1:1. Chalcones with hydrogen or electron withdrawing substituents (Cl, CHO, COOCH(3), and NO(2)) formed aurones 3. Synthesized isoflavones 2 and aurones 3 were evaluated for their inhibitory activity against interleukin-5. Among them, 5-(cyclohexylmethoxy)-3-(3,4,5-trimethoxyphenyl)-4H-chromen-4-one (2h, >100% inhibition at 50 microM, IC(50)=6.1 microM) gave most potent activity. All the aurones 3 were inactive.

The role of alkoxy group on the A ring of isoflavones in the inhibition of interleukin-5.

Novel isoflavones were found to be potent inhibitors of interleukin-5 (Il-5). 5-Benzyloxy-3-(4-hydroxyphenyl)-4H-chromen-4-one (2a, 87.8% inhibition at 50 microM, IC50 = 15.3 microM) was initially identified as a potent inhibitor of IL-5. Its activity was comparable to that of budesonide or sophoricoside (1a). The benzyloxy group appeared to be critical for the enhancement of the IL-5 inhibitory activity. To identify the role of this hydrophobic moiety, 5-cyclohexylmethoxy (2d), 7-cyclohexylmethoxy (2e), 5-cyclohexylethoxy (2f), 5-cyclohexylpropoxy (2g), 5-(2-methylpropoxy) (2h), 5-(3-methylbutoxy) (2i), 5-(4-methylpentoxy) (2j) and 5-(2-ethylbutoxy) (2k) analogs were prepared and tested for their effects on the bioactivity of IL-5. Compounds 2d (IC50 = 5.8 microM), 2e (IC50 = 4.0 microM) and 2j (IC50 = 7.2 microM) exhibited the most potent activities. Considering the cLog P values of compounds 2 and the different three dimensional structures of 2d and 2e, the alkoxy group on ring A contributed to their cell permeability for the enhancement of activity, rather than playing a role in the ligand motif binding to the receptor. The optimum alkoxy group should be one that provides cLog P values of compounds 2 in the range of 4.13 to 4.39.

Structural requirement of chalcones for the inhibitory activity of interleukin-5.

Novel chalcones were found as potent inhibitors of interleukin (IL)-5. 1-(2-Benzyloxy-6-hydroxyphenyl)-3-(4-hydroxyphenyl)-2-propen-1-one (2b, 78.8% inhibition at 50microM, IC(50)=25.3microM) was initially identified as a potent inhibitor of IL-5. This shows the compatible activity with budesonide or sophoricoside. To identify structural requirements, 26 chalcones were prepared and their inhibitory activities were tested against IL-5. Among them, compound 4-[(E)-3-(2-cyclohexylmethoxy-6-hydroxyphenyl)-3-oxoprop-1-enyl]benzenesulfonamide (2w, 99.5% inhibition at 50microM, IC(50)=1.8microM) shows the most potent activity. The important structural requirements of these chalcone analogs exhibiting the inhibitory activity against IL-5 were recognized as the following. (1) The hydrophobic group such as benzyloxy or cyclohexylmethoxy at 6-position of A ring is necessary. (2) The existence of phenolic hydroxyl at 6-position of A ring is critical. (3) Propenone unit as alpha,beta-unsaturated ketone is essential. (4) Electron withdrawing groups with hydrogen acceptor property at 4-position of B ring enhance the activity and quantitative structure-activity relationship of 2 regarding these substituents was determined.

The role of the hydrophobic group on ring A of chalcones in the inhibition of interleukin-5.

Novel chalcones were found as potent inhibitors of interleukin-5 (11-5). 1-(6-Benzyloxy-2-hydroxyphenyl)-3-(4-hydroxyphenyl)propenone (2a, 78.8% inhibition at 50 microM, IC50 = 25.3 microM) was initially identified as a potent inhibitor of IL-5. This activity is comparable to that of budesonide or sophoricoside (1a). The benzyloxy group appears to be critical for the enhancement of the IL-5 inhibitory activity. To identify the role of this hydrophobic moiety, cyclohexyloxy (2d), cyclohexylmethoxy (2c), cyclohexylethoxy (2e), cyclohexylpropoxy (2f), 2-methylpropoxy (2g), 3-methylbutoxy (2h), 4-methylpentoxy (2i), and 2-ethylbutoxy (2j) analogs were prepared and tested for their effects on IL-5 bioactivity. Compounds 2c (IC50 = 12.6 microM), 2d (IC50 = 12.2 microM), and 2i (IC50 = 12.3 microM) exhibited the most potent activity. Considering the cLog P values of 2, the alkoxy group contributes to the cell permeability of 2 for the enhancement of activity, rather than playing a role in ligand motif binding to the receptor. The optimum alkoxy group in ring A of 2 should be one that provides the cLog P of 2 in the range of 4.22 to 4.67.