画像解析による大腸癌早期診断の可能性と限界

Diagnosis of Early Colorectal Cancer Based on Computer Analysis of Magnifying Endoscopic Image

要旨　早期大腸癌の定量的鑑別診断をするため，電子内視鏡を用いて大腸隆起性病変（化生性ポリープ5例，腺管腺腫23例，早期癌7例，進行癌12例）の拡大観察を行い，表面パターンを抽出して「幅」，「面積」，「伸長度」，「不整度」，「拡張度」の5要素，個々の腺ロパターンから「輝度」の画像解析を行った．その結果，「面積」，「不整度」，「拡張度」，「輝度」の解析において早期癌は進行癌と腺管腺腫の中間値を示し，鑑別診断の可能性が示唆された．今後は腺口の輪郭をさらに正確に描出できる高い倍率を有するスコープを用いること，病変との距離を安定させるため操作性を良くすること，ノイズを除去することなどの改良を加えることによって，より精度の高い早期癌の識別，更にm癌とsm癌の鑑別も可能になると期待された．

　To introduce a quantitative approach in the diagnosis of early colorectal cancer, we conducted magnifying observation for colorectal elevated lesions and analyzed the images by computer. The subjects included 5 cases with metaplastic polyp, 23 with tubular adenoma, 7 with early cancer and 12 with advanced cancer. After extraction of surface patterns, 6 indices were calculated: "width", "area", "elongation", "irregularity", "dilatation" and "brightness". The results were that the indices including "area", "irregularity", "dilatation" and "brightness" of early cancers were valued between those of advanced cancers and of tubular adenomas respectively. These results suggested the possibility of a quantitative approach in the diagnosis of early cancer. For more precise diagnosis, the following improvements are considered to be needed: (1) use of instrument with higher magnifying power, (2) mechanical device to keep the distance between the scope and object constant, (3) elimination of noise.