Exploring artificial neural network combined with laser-induced auto-fluorescence technology for noninvasive in vivo upper gastrointestinal tract cancer early diagnosis

In this study, a laser-induced auto-fluorescence (LIAF) system combined with the artificial neural network (ANN) algorithm is developed for early detection of human upper gastrointestinal tract carcinoma in vivo, through investigating the LIAF spectrum characteristics of the normal mucosa layer and the changes concerning an abnormal surface. Of the 44 participating patients, 41 underwent biopsy at the abnormal surface area at endoscopy. The ANN is employed to differentiate the LIAF data obtained from the normal and carcinoma patients (according to biopsy pathology diagnosis). The LIAF spectrum between 500 and 700 nm is selected and normalized. One data point is selected every 10 nm. A feed-forward back-propagation network with 2 hidden layers is constructed and trained. To evaluate the performance of ANN, 10 normal and 10 carcinoma data sets are tested with the trained ANN. 100% of the carcinoma data are very close to −1 (desired), 80% of the normal surface is very close to 1 (desired), and 20% return values around −0.28. Previous works on this type of ANN suggested a threshold of −0.5. As a result, all normal data are successful and the carcinoma cases are accurately classified and diagnosed. In conclusion, the LIAF technology combined with ANN diagnosis is more accurate.