A SEMIAUTOMATIC PROCEDURE FOR EDGE-DETECTION OF IN-VIVO PULMONARY MICROVESSELS AND INTERSTITIAL SPACE

Objective: To develop an algorithm to detect tile edges between lung t issue, perivascular interstitium, and microvessel using digital proces sing of in vivo microscopic images of lung surfacer. Methods: A numeri cal technique was developed to identify three different regions (namel y, pulmonary microvessel, perivascular interstitium: and lung tissue) based on their corresponding gray level distributions. We present a th eoretical demonstration of the method and a semiautomatic procedure th at, once tile edges are detected, determines microvascular diameters a nd perivascular interstitium thickness. Results: Microvessel diameters and perivascular interstitium thickness were calculated for precapill ary arteriolar branching (40 to 140 mu m) and saved in an ASCII file. Conclusions: We proved that the maximum value of the moving variance i s useful to detect the edge between two adjacent regions whose gray le vel distributions satisfy the condition: \sigma(Y)(2)-sigma(X)(2)\ les s than or equal to (mu(X)-mu(r))(2), where mu(X),mu(Y),sigma(X)(2),sig ma(Y)(2) are the statistical moments of the two regions X and Y. Moreo ver, when the legions have similar means, the above condition is not m et; but tile edge between them can be detected by tile maximum of the moving variance error.