Computational Digital Pathology Lab


Our vision and objective is to tailor-make and prototype computational medical image analysis solutions, including pathological images, such as Hematoxylin-Eosin (H&E) staining and Immunohistochemistry (IHC) staining image analysis, and general bio-image quantification methods. We also focus on developing novel image analysis algorithms for other tricky problems in the field.

Our team focuses on the following topics in pathological/biomedical image analysis:

  • Developing novel computational image processing algorithms for basic scientific research, biomedical and drug discovery applications
  • Developing practical industry-ready software packages for automated image/video analysis to improve clinical practice
  • Building mathematic models for the classification/prediction using pathologist annotation & Histopathological image databases
  • Developing machine learning, AI-based, and deep learning solutions for big biological/biomedical imaging data

In recent years, with the advent of digital image scanners and artificial intelligence, histology laboratories and standards are undergoing fundamental changes. In order to promote the adoption of modern digital solutions in daily pathologists’ practice, automated quality assessment tools are essential. The evaluation of image quality is an active area of research. The perceptual image quality estimator is needed to calculate the image's non-reference image quality score. However, compared with histological images, this method shows higher efficiency on natural images. Some tools have previously been developed for histological images, but the "quality assessment" is limited to “out-of-focus” in H&E images. Therefore, for any imaging method related to histology, we have developed a dedicated solution specifically related to nuclear image quality assessment. At this stage, this software package has been validated by digital image scanner manufacturers.



We have developed and proposed a digital pathology annotation platform to realize the concept of interaction between pathologists and AI. The platform provides a highly interactive visual viewer and annotation tools to perform annotation tasks. AiHistoNote is specially developed for human elements. It has a modern appearance and style that can improve readability and navigation. Therefore, it can enhance intuitive interaction and effective workflow. At the same time, AiHistoNote is a computer-aided diagnosis platform that uses the interaction between pathologists and AI models to improve the efficiency and consistency of diagnosis. The machine can alert the pathologist and attract it to the area of interest, and determine the tumor grades, which the pathologist can accept or modify. Currently, senior pathologists at the National University Hospital of Singapore (NUHS) have used the platform to test and perform annotation tasks..


We develop a deep learning based model for automated Gleason Grading in prostate histopathological images. By learning from pathologists’ annotations, the AI model learns how to differentiate Gleason Pattern 3, Gleason Pattern 4, Gleason Pattern 5, normal glands, stroma and other structures. The model is able to detect malignant region at different scale and alert pathologists to some small malignant regions. .It is also applicable to images scanned by scanners from different brands (Akoya Biosciences, Olympus, Zeiss, KFBio, Leica, etc.). The automatic detection increases the accuracy, efficiency and consistency of the Gleason Grading and quantitative tumour burden assessment.

The preliminary model we build is capable of incremental learning, which means the model performance will improve gradually with more new training data. Through the interaction between the AI model and pathologists, we can collect new annotated data more efficiently. .

BII_ciid-cpdl-figure5 Mitotic count is a crucial clinical marker for diagnosis of breast cancer. We develop pipeline for semi-auto mitotic figures detection in breast cancer histopathological images and validated on three different data sets: SGH dataset, ICPR2012 and ICPR2014. We separate the Haematoxylin and eosin channel by colour deconvolution and analyse the intensity features of mitotic figure in Hematoxylin channel since Hematoxylin basically stains only the nuclei, where mitosis occurs and cells divide. We select three most discriminative features and train a Naïve Bayes model to differentiate mitotic figures and background. By applying the trained model, we can filter out the regions that don’t have any mitotic figure and only keep the rest (~15% of the whole area) for pathologists’ examination. The output of the model assists the pathologists to locate the ‘mitosis hot spots’ when they count the number of mitotic figures. It is applicable to images of different sizes, including TMA images.

BII_ciid-cpdl-figure6 We have developed an automated image processing pipeline to detect and quantify multiple immunophenotypes in histological prostate tissue samples. Our pipeline integrates powerful image processing and machine learning methods to analyse histopathological tissue whole slice images (WSI). We have stratified the patient cohort (N = 156) based on the patient Gleason score (GS, primary and secondary Gleason grade) annotated by the pathologists. We cut each patient’s tissue sample into 4 consecutive slices. The first piece is stained with H&E so that the pathologist can evaluate. The subsequent 3 sections were stained with UltiMapper I/O PD-L1, UltiMapper I/O PD-1 and UltiMapper I/O T-act kits. We use pipelines to quantify the immunophenotype along the four consecutive slices in terms of cell density (#cells/mm2). Then, we analysed the statistical significance of these immunophenotypes between patient groups and different tumour regions. For a variety of different patient groups, we used multiple immunophenotypes to obtain significant results, p <0.05. Our results combine basic overall cell density and neighbouring cell density, paving the way for more complex quantitative cell-cell interactions that can be used in the clinical environment for cell distribution pipelines.

BII_ciid-cpdl-figure7 Cervical cancer can be effectively treated if diagnosed early. Artificial intelligence has the potential to reduce pathologists’ workload, improve diagnostic speed and maintain high accuracy for cervical cytology screening. To improve diagnostic accuracy and speed, we developed an AI assistive diagnosis system to classify cervical liquid-based thin-layer cell smears according to the TBS criteria.

In this study, we trained our models with >81,000 samples retrospectively. We use YOLOv3 to develop the target detection model (mAP, Mean Average Precision, up to 0.8233), Xception algorithm and Patch-based classification to boost the detection target classification model (accuracy rates of 0.968 and 0.9192, respectively), U-net to segment the nucleus and accurately measure the gray value of the nucleus ( Mean intersection, mIOU reaches 0.8356), We integrated XGBoost and logical trees together with the above deep learning models of the optimized fetures given by the deep learning process, formulated a complete cervical liquid-based cytology smear TBS classification and diagnosis pipeline, and successfully established an artificial intelligence-assisted diagnosis system. The optimized system was validated with >34,000 prospectively samples collected from 11 medical Institutions.


 Principal Investigator  YU Weimiao   |    [View Bio]  
 Post-Doctoral Research Fellow ONG KokHaur  
 Research Officer HUO XinMi

Selected Publications