To localize a tumor in the bladder, doctors routinely perform a cystoscopy, a procedure that involves inserting a lensed, lighted tube into a patient to enable their search for potentially cancerous sites. Cancer can easily be missed because doctors rely only on a narrow optical tube and their eyesight for visual analysis. Fluorescent labeling and other markers have been suggested, but these methods only stain for expected changes. Raman spectroscopy can produce a complete biochemical profile of the tissue and can be integrated into a cystoscope through fiber optics. For the initial stage of bladder cancer, which is particularly difficult to detect with cystoscopy, Raman spectroscopy might be more sensitive. Doctors use cystoscopes to search for bladder cancer, but the view is limited, so researchers want to integrate them with Raman spectroscopy equipment to provide biochemical information during the procedure. After performing cystoscopy, doctors biopsy suspicious tissue and view it under the microscope, in contrast to Raman spectroscopy, which can examine tissue nondestructively. However, analysis of Raman spectra is a painstaking process — time consuming for both doctors and patients. So, Bas W.D. de Jong and colleagues at Erasmus Medical Center in Rotterdam, the Netherlands, have programmed software to use algorithms that enable automatic analysis of Raman spectra. The investigators examined bladder tissue from 15 patients with a Leica microscope equipped with an 80x Olympus objective with a 0.75 NA because, as de Jong explained, it enabled near-IR measurements with low noise. For an 845-nm excitation source, the researchers employed a tunable Spectra-Physics Ti:sapphire laser that was pumped with a Coherent argon-ion laser. Raman scatter passed through a chevron filter, which de Jong said was more stable than a holographic filter, and into a Renishaw spectrometer. The scientists focused the beam 2 μm below the tissue surface with a spot size of 2 and collected the signal in 20 s. They programmed the software to group biochemical data into broad clusters and to look for subtle differences between them. To test the accuracy of those algorithms, they instructed the software to exclude one patient and to use the remaining patient data to predict whether that patient had cancer; the process was repeated for each patient. From a pathologist’s independent histological investigation, the researchers knew which tissues represented cancer and which did not. They used that information to determine whether the Raman spectroscopy and their algorithms yielded the same results. They found that bladder tumor tissue expressed more lipids, glycogen and nucleic acids than nontumor tissue and had a higher protein content. However, nontumor tissue contained higher amounts of certain lipids such as arachidonic acid and stearic acid. The researchers also detected β-carotene in one tumor tissue sample. Anecdotally, carotene is believed to combat bladder cancer, so the researchers assumed that the patient purposefully ingested it. This finding suggested that Raman spectroscopy could be used to examine whether drugs are absorbed at their target sites. Overall, the algorithms correctly identified 94 percent of the tissue. Considering only false-negatives and false-positives, respectively, the algorithms were 94 and 92 percent accurate. The scientists noted that the cutoffs for false-positive and false-negative prediction could be adjusted to meet a doctor’s preference. Inflamed tissue caused false-positive results because it exhibits high RNA and DNA expression, as does tumor tissue. However, de Jong said that the differences between tumor tissue and inflamed tissue will reveal themselves as they test more samples and input their spectra into a database. Analytical Chemistry, Nov. 15, 2006, pp. 7761-7769.