Evaluation of esophageal cytology using a neural net-based interactive scanning system (the PAPNET system): its...

A neural net-based, semiautomated, interactive computerized cell analysis system (The PAPNET system, Neuromedical Systems, Suffern, NY) was used to examine cells from 138 esophageal smears obtained by lavage, brushings, or balloon from as many patients. From each smear, trained human observers examined 128 cell images selected by the machine. Abnormal cells were identified in all 35 patients with cancer, whether esophageal, gastric, oral, or metastatic. Further, in 11 smears, the displayed images allowed the recognition of effects of radiotherapy and, in 14 smears, the diagnosis of a specific tumor type, such as squamous cell carcinoma (8 patients) or adenocarcinoma (6 patients). In 3 additional cases, the diagnosis of "carcinoma, not further specified," was established. One case of esophageal carcinoma in situ, not previously recognized on a smear or in the biopsy specimen, and one case of gastric adenocarcinoma, not recognized in the smear, were identified in PAPNET-generated images. The possible application of the apparatus to the triage of smears and population screening for esophageal and gastric carcinoma precursors is discussed.