CASUS Institute Seminar, Dr. Harry Horsley, Senior Research Associate, Bladder Infection and Immunity Group (BIIG), University College London (UCL), United Kingdom

Harry is the basic sciences lead (PI) of the Bladder Infection and Immunity Group (BIIG), based in the Department of Renal Medicine, Division of Medicine, UCL. He has studied urinary tract infections (UTI), from the pathophysiology of intracellular infection to the exploration of novel treatments strategies for over 12 years. He completed a few years as a postdoctoral research associate and deputy head of the Centre for Urological Biology before receiving a generous research donation to form and head the BIIG. Currently, in addition to his research and leadership role, he founded and runs a ~£1,000,000 state-of-the-art biological imaging suit at the Department of Renal Medicine.

Urinary tract infection (UTI) is a significant cause of morbidity, ranking as one of the most prevalent infectious diseases worldwide. Acute UTI is not diagnostically challenging, as the rapid onset of urinary frequency and dysuria are clear indicators of the pathology. However, recalcitrant chronic bladder pain and recurrent cystitis in patients with negative urinalysis present a worrying management problem. These patients fail treatment protocols for uncomplicated urinary tract infection, overactive bladder and chronic bladder pain.

There is increasing evidence that patients presenting with lower urinary tract symptoms (LUTS) such as urinary frequency, urgency and bladder pain, may harbor a UTI despite negative tests. One of the problems lies in the handling and processing of urine specimens and the diagnostic criteria used in culture-based diagnosis. A significant body of literature points to the inherent flaws of quantitative urinary microbiological analysis. Dipstick urinalysis performs equally poorly, hampered by insensitivity and spectrum bias. These tests cannot reliably exclude UTI and do not take into account differences in bacterial strain virulence, host genetic variability, intracellular bacterial reservoirs, or even urine dilution due to high fluid intake before the test.

Current gold-standard diagnostic tests for UTI have been widely discredited in numerous high-profile publications. Currently, the best diagnostic test available is the use of bright field microscopy to identify and innumerate white blood cells (pyuria) and shed urothelial cells in patient urine samples. This is a highly skilled and time-expensive undertaking in a clinical setting and suffers from unacceptable inter-observer variability. In collaboration with Dr. Artur Yakimovich (CASUS) Harry’s lab has combined high-content screening microscopy in conjunction with deep-learning (DL) convolutional neural networks to begin fully automating this process. To date over 28,500 images from the urine of nearly 2,000 recurrent UTI patients over multiple clinic visits, and 40 premenopausal healthy controls have been longitudinally acquired. 10,000 of these images have been ‘annotated’ and the morphometric data of over 600,000 urinary cells clustered using unsupervised machine learning (ML).

To obtain unbiased annotations suitable for their supervised ML model, Harry employed a semi-supervised ML approach accompanied by active leaning. In this approach, all the foreground objects in the raw images are segmented using a pixel classifier (random forest algorithm) trained via active learning. Next, these objects were clustered using unsupervised ML techniques based on morphometric data. These clusters will, in the near future, be used to pretrain a multi-class DL neural network on object detection and segmentation tasks.

This system has the potential to revolutionize diagnosis and patient management. Furthermore, if found successful, this fully trained platform could potentially be translated into a user-friendly standalone device.

The event is organized with a videoconferencing tool by Zoom Inc. If you want to join the talk remotely, please ask for the login details via