Express Healthcare

Artificial Intelligence and diagnosis of prostate cancer

0 256

Chaith Kondragunta, CEO, AIRA Matrix talks about AI based diagnosis of prostate cancer

Prostate cancer is the second most frequent cancer and the fifth leading cause of cancer death among men. The disease exhibits a broad spectrum in patients, ranging from indolent and slow growing (needing only watchful waiting) to aggressive and lethal disease (needing immediate therapeutic intervention). It thus warrants tailored treatment regimens. In addition, under diagnosis can lead to increased mortality due to the disease while over diagnosis can lead to unnecessary treatment resulting in morbidity due to adverse effects. Hence for effective treatment of prostate cancer, accurate disease stratification and tailored therapy decisions are as important as early disease detection and diagnosis.

Unfortunately, though early detection of prostate cancer is possible, early discrimination is not; which creates uncertainty about disease progression and treatment decisions. Hence in addition to sensitive and reasonably specific screening methods for early disease detection and diagnosis, techniques for improved disease stratification and prognostication that help institute appropriate therapy regimens are the cornerstone of the prostate cancer care pathway. At AIRA Matrix, we aim to positively impact prostate cancer outcomes by applying AI-based solutions for diagnosis, prognosis and prediction across the care pathway – from screening to post-surgical follow ups.

The Gleason score is an important histological parameter for stratifying the cancer into different grade groups and to aid therapeutic decisions. This is currently determined by specialised pathologists, through microscopic examination of the biopsy tissue. Widely noted inter-observer variability in Gleason grading, particularly at key clinical decision points such as Gleason Grade Group 1 (which may require active surveillance only) versus Gleason Grade Group 2 (which may require therapeutic intervention), compounded by scarcity of subspecialty expertise required to achieve optimal grading precision, often leads to inconsistent disease stratification.

These type of issues are extremely suitable for the application of deep learning techniques. We have developed a novel deep learning based solution for pixel-level semantic segmentation. This closely follows the approach an experienced pathologist applies and provides for superior performance over current approaches. This approach also reduces inter-observer variability in reporting of histopathology images – and has led to improved accuracy, precision and speed in Gleason’s scoring.  We believe these improvements will enable better prognostication of prostate cancer and help guide selection of appropriate therapy regimens.

Beyond Gleason grading, we are currently working on solutions that predict the course, spread and aggressiveness of the cancer. Our goal is to improve screening, prognostication, therapy planning and follow up of prostate cancers – with better outcomes for the patient as well as care-givers.

The prostate cancer care pathway spans biopsy and other investigations for early disease detection and diagnosis, disease stratification using biochemical, radiological and histopathological parameters to grade and stage the disease to decide on therapy options like watchful waiting, active surveillance, radical surgical intervention, chemotherapy, radiotherapy, as well as careful follow up of treated patients, grouped into these different therapy regimen groups.

Our products and solutions are aimed to act at every step of the prostate cancer care pathway and, make it more effective and outcome oriented. In the screening process, our triaging solution helps in rapid screening of prostate needle core biopsies to detect and diagnose malignancy. This algorithm can act in a two-fold manner, by screening and triaging biopsy sections to point out critical areas to the pathologist, or as a second read, to identify tumour areas that may have been over looked by the pathologist. The dramatic increase in the number of CNBs reviewed per case over the past decade has resulted in increased workload on the pathologist – potentially resulting in delayed cancer diagnoses and possible diagnostic errors due to fatigue. Our solution looks to ease this burden on the pathologist.

Our solutions also help in the risk stratification steps by providing gland wise annotation on each core biopsy, and automatically computing the Gleason grade and score. The goal is to reduce subjectivity and inter-observer variability.

Finally, we are looking to utilise information from the histopathology gold standard to augment other non-invasive modalities like radiology and molecular testing. For example: Multi-parametric Magnetic Resonance Imaging (mpMRI) is an important diagnostic modality. It is used in primary diagnosis, risk stratification and therapy of prostate cancer patients. However, mpMRI substantially underestimates size, extent and tumour volumes in comparison with histopathology. We are working on 3D reconstruction of tumour specimens by co-registration of mpMRI and radical prostatectomy images. We hope to improve staging, focal therapy planning and follow ups via these less invasive procedures.

We believe these solutions will enable appropriate and timely therapy options and improve healthcare outcomes across the prostate care pathway.

- Advertisement -

Leave A Reply

Your email address will not be published.