The invasive and expensive diagnosis process of bladder cancer, which is one of the most common and aggressive cancers in the United States, may be soon helped by a novel non-invasive diagnostic method thanks to advances in machine learning research at the San Diego Supercomputer Center (SDSC), Moores Cancer Center, and CureMatch Incorporated.
Research scientists Igor Tsigelny and Valentina Kouznetsova have been working on the development of a machine-learning (ML) model that looks at a patient’s metabolites and their chemical descriptors.
The model accurately classifies the stages of bladder cancer in a patient, according to the researchers.
Tsigelny is the lead author on a recently published study in the Metabolomics journal called ‘Recognition of Early and Late Stages of Bladder Cancer using Metabolites and Machine Learning’.
When a patient experiences early symptoms of bladder cancer (e.g., blood in urine, pain during urination, etc.), the current method of diagnosis is often a painful, invasive series of tests.
“From my point of view, it can be very easy for patients just give a sample of urine and our ML system can produce a “red flag” analysis result telling them to go immediately to an oncologist for testing,” said Tsigelny.
“We believe that a lot of early stages and even more advanced stages of bladder cancer go untreated because patients don’t pay attention to mediate pain signals from the body, and may be thinking that there are less dangerous problems causing the symptoms.
Our machine learning model uses metabolites and corresponding genes to determine if a patient has bladder cancer and if so, at what stage.”
More than 81,000 Americans were diagnosed with bladder cancer in 2018 and of those, more than 17,000 died from the condition, according to statistics from the American Cancer Society.
“The goal of this research is to lower that number and we believe that machine learning models can help us do that,” said Kouznetsova.
“Using a variety of computational tools, we studied pathways related to different stages of bladder cancer that can be used for diagnostics and monitoring of cancer progression.”
The researchers trained the software – called multi-layer perceptron or MLP – with the data of urine metabolites of the patients with the different stages of the disease.
Each stage has its own profile of metabolites.
“MLP analyzes the chemical descriptor of the sets of metabolites related to each stage of cancer and creates AI models of these profiles,” explained Kouznetsova.
Tsigelny, along with his work at UC San Diego, is the chief science officer and co-founder of CureMatch, which provides decision support for doctors in personalized cancer medicine. SDSC Director Michael Norman is a member of the CureMatch Advisory Board.
“With 4.5 million possibilities to combine around 300 FDA-approved cancer drugs, CureMatch targets multiple cancer mutations at the same time and determines the best combination treatment for each patient,” explained Tsigelny.
“While this study is not related to the current tasks of CureMatch, it may become so in the future.”
Eden Romm, a bioinformatics specialist at CureMatch, and SDSC Research Experience for High School Students (REHS) participants Elliot Kim and Alan Zhu also participated in this study.
Current Diagnostic Tools for Bladder Cancer
Bladder cancer is the fourth most common cancer in men and the eighth most common cancer in women in the Western world [1].
Worldwide, it is the seventh most common cancer in men [2,3]. It occurs four times more in men than in women.
Current diagnostic tools to detect bladder cancer are cystoscopy and cytology.
Cystoscopy is an effective but invasive tool to detect bladder cancer tumors.
Moreover, it has a low sensitivity for carcinoma in situ (Tis) and tumors can still be missed because effectiveness is operator-dependent, especially for the detection of recurrence [4].
Sensitivity and specificity range from 62 to 84% and 43 to 98%, respectively, depending on the type, stage, and grading of the tumor [5].
In addition, pain during urination (50%), urinary frequency (37%), visible hematuria (19%), and infection (3%) are experienced relatively frequently after flexible cystoscopy [6,7].
Urine cytology is a non-invasive diagnostic method used in clinical practice where voided or instrumented urine is examined for exfoliated cancer cells.
The overall sensitivity ranges from 28 to 100%, with a median of 44%. It has a high sensitivity for high-grade tumors, but low sensitivity for low-grade tumors, ranging only from 4% to 31% [8,9,10].
Cytology is useful, particularly as an adjunct to cystoscopy, when a high-grade malignancy is present.
A positive cytology indicates a urothelial tumor anywhere in the urinary tract. However, negative cytology does not exclude the presence of a tumor.
Cytological interpretation is also user dependent and can be hampered, for example, by low cellular yield, urinary tract infections, and stones [11].
Urine-Based Biomarkers for Bladder Cancer
The use of urine-based biomarkers to detect bladder cancer seems to be an attractive alternative.
Konety et al. (2006) defined the ideal bladder cancer biomarker as an objective, non-invasive, easily interpreted marker, possessing high sensitivity and specificity [12].
Urinary biomarkers are in direct contact with the bladder and can come in a variety of forms such as proteins, metabolites, DNA, different types of RNA, and single nucleotide polymorphisms (SNPs).
Presence or variations in expression of those molecules could be linked to bladder cancer [13]. In the following section, Food and Drug Administration (FDA)-approved urine tests for bladder cancer are described.
Table 1 gives an overview of these FDA-approved tests, the detected biomarkers and the assay type, as well as their sensitivity and specificity.
Table 1
Commercially available FDA-approved tests for bladder cancer. The biomarker and assay type of the tests are included in the table. The mean and range (between brackets) of the overall sensitivity and specificity are also shown. Adapted from [12,14,23,24,29].
| Test | Biomarker | Assay Type | Sensitivity (%) | Specificity (%) |
|---|---|---|---|---|
| NMP22® BC test | NMP-22 | Sandwich immunoassay | 69 (26–100) | 77 (41–92) |
| NMP22® BladderChek® | NMP-22 | Sandwich immunoassay | 58 (51–85) | 88 (77–96) |
| BTA stat® | Complement factor H-related protein | Colorimetric immunoassay | 64 (29–83) | 77 (56–86) |
| BTA TRAK® | Complement factor H-related protein | Sandwich immunoassay | 65 (53–91) | 74 (28–83) |
| ImmunoCyt™ | Carcinoembryonic antigen and 2 mucins | Immunofluorescence cytology | 78 (52–100) | 78 (63–79) |
| UroVysion™ | Aneuploidy of chromosomes 3, 7, 17 and loss of 9p21 locus | Multitarget FISH | 63 (30–86) | 87 (63–95) |
More information: Valentina L. Kouznetsova et al. Recognition of early and late stages of bladder cancer using metabolites and machine learning, Metabolomics (2019). DOI: 10.1007/s11306-019-1555-9
Provided by University of California – San Diego
