AI Can Improve Autosomal Dominant Polycystic Kidney Disease Diagnosis & Management – Physician’s Weekly

Anish Raj, MSc

Anish Raj, MSc, Computer-Assisted Clinical Medicine, Mannheim Institute for Intelligent Systems in Medicine, Medical Faculty Mannheim, Heidelberg University, Germany

Evidence indicates that because autosomal dominant polycystic kidney disease (ADPKD) is characterized by the gradual development and growth of cysts on the kidneys that can compromise renal function, monitoring this cystic progression is vital to optimal care. Cyst enlargement eventually leads to end-stage renal disease (ESRD) with complete renal failure.

Early detection is important but can be difficult. “One of the biggest challenges is to detect [ADPKD] at an early age,” says Anish Raj, MSc. “In a child, the cysts might be too small to be noticeable, even in MRI images. Early detection can help delay the onset of kidney failure by a couple years.”

Accurate Total Kidney Volume

Once diagnosed, total kidney volume (TKV) is an important biomarker for determining renal function in patients and can be used as a risk predictor of disease development, explains Raj. TKV is determined by segmenting the kidneys from the MRI volumes. However, even for an experienced practitioner, manual segmentation can be time-consuming and prone to observer variability.

These obstacles to accurate TKV calculation may be overcome through technological advancements. Raj’s expertise in computer-assisted clinical medicine provides a unique insight into achieving optimal ADPKD monitoring: “For practitioners with a background in programming and machine learning, achieving such monitoring is likely to be straightforward,” he adds. “However, most physicians, I believe, are not invested heavily in the technical side of it, so for them, it might be difficult. Nonetheless, the techniques I have implemented to do this are quite readily available as open source and do not require years of coding experience. It is possible that someone can create an application out of it, and then physicians can use that application to apply to their data.”

Deep-Learning Approach

Raj is referring to a deep-learning approach to handling the data that monitors and clarifies the progression of ADPKD. Deep learning is a branch of machine learning that has shown some promise in the advancement of medical care. It assists in data classification, novel disease phenotyping, and complex decision-making.

When asked about the current capabilities available to gather data and monitor the progression of ADPKD, Raj explains, “At the moment, automated segmentation can provide accurate results for many cases, except for outliers in the data. It is important to train AI [artificial intelligence] models with as diverse a data pool as possible, but in the medical domain, there is always a scarcity of data due to data privacy and the cost of patient data acquisition.”

For a study published in Diagnostics, Raj and colleagues showed that combining the cosine loss function (a method of evaluating how well an algorithm models a dataset) and sharp awareness minimization (a procedure that aims to improve model generalization by simultaneously minimizing loss value and loss sharpness) could reach a high level of accuracy compared with manual segmentation of TKV, which is what many practitioners use.

Promising Future

According to Raj, “The next step for the future would be to have a dataset that can help train AI models so they can be applied in more places. Furthermore, these automated segmentations are not perfect, so I would always advise a physician to check them and do manual adjustments when needed. For improving TKV estimation, it would be better if we can get CT and MRI from the same patient and then train the algorithm on both of the imaging modalities to see how they vary in their TKV estimation. If there is only a slight difference, then we can take the average from both and get a more precise estimation.”

Advancements in segmentation accuracy will help treat patients with ADPKD. Raj is very hopeful, “The algorithm certainly helps the physicians by allowing them to not spend more time on kidney segmentation and, hence, TKV estimation. The TKV estimation can provide information on how the patient’s ADPKD condition is now and in the near future.”

Spread the love

Leave a Reply

Your email address will not be published.