Development of a system which automatically acquires optimal discrete-valued attributes by dividing and grouping continuous-valued attributes to assist clinical decision making in radiotherapy.
Hiroko Kou, Hajime Harauchi, Hodaka Numasaki, Yu Kumazaki, Yasuhiro Okura, Akihiro Takemura, Takashi Kondou, Masatoshi Ishibashi, Kuniyuki Hidaka, Tokuo Umeda, Kiyofumi Haneda, Kiyonari Inamura · PubMed · 2003
The purposes of this study were first to develop a system which statistically tests results of radiotherapy and which automatically acquires an optimal discrete-valued attribute by dividing and grouping continuous-valued attributes, and second to find the optimal range of values such as tumor dose by taking account of the conditions and statistics in ROGAD (Radiation Oncology Greater Area Database), a multi-institutional database in Japan. Our ultimate goal is to assist clinical decision making for every patient. In this research, two algorithms for acquiring a boundary value were developed without detecting false boundaries or accidental errors of acquired boundaries. The resolution of detected discrete-valued attributes and speed of convergence were confirmed to be practical. The optimal range of given tumor dose with the best reaction and with the fewest complications is expected to be clarified.