Quality-assured Efficient Engineering of Feedforward Neural Networks (QUEEN)

K. Hensler; T. Waschulzik; S. P. Mönig; K. Maruyama; A. H. Hölscher; E. Bollschweiler

doi:10.1055/s-0038-1634021

Methods of Information in Medicine, Table of Contents

Methods Inf Med 2005; 44(05): 647-654
DOI: 10.1055/s-0038-1634021

Original Article

Schattauer GmbH

Quality-assured Efficient Engineering of Feedforward Neural Networks (QUEEN)

Pretherapeutic Estimation of Lymph Node Status in Patients with Gastric Carcinoma

K. Hensler

¹Department of Visceral and Vascular Surgery, University of Cologne, Cologne, Germany

,

T. Waschulzik

³Carl Zeiss Vision GmbH, Hallbergmoos near Munich, Germany

,

S. P. Mönig

¹Department of Visceral and Vascular Surgery, University of Cologne, Cologne, Germany

,

K. Maruyama

²Department of Surgical Oncology, University of Health and Welfare Sanna Hospital, Tokyo, Japan

,

A. H. Hölscher

¹Department of Visceral and Vascular Surgery, University of Cologne, Cologne, Germany

,

E. Bollschweiler

¹Department of Visceral and Vascular Surgery, University of Cologne, Cologne, Germany

› Author Affiliations

Abstract

PDF Download

Summary

Objectives: Lymph node metastasis (LNM) is an important prognostic indicator in patients with gastric carcinoma. However, the methods that have been established for preoperative diagnosis of LNM show insufficient accuracy.

Methods: This study describes the use of the QUality Assured Efficient Engineering of Feedforward Neural Networks with Supervised Learning (QUEEN) technique to attempt optimization of the preoperative diagnosis of lymph node metastasis in patients with gastric carcinoma. The results were compared with the Maruyama Diagnostic System (MDS) for preoperative prediction of LNM, established at the National Cancer Center in Tokyo.

Results: QUEEN is able to extract predictive variables from a case-based database. The combination of a development method, a special type of neural network and the corresponding encoding yielded an accuracy of 72.73%, which is notably higher than that of the MDS. Our system produced a nearly ten per cent higher sensitivity and around eighteen per cent higher specificity than MDS.

Conclusion: Our results show that QUEEN is a reasonable method for the development of ANNs. We used the QUEEN system for prediction of LNM in gastric cancer. This system may allow more meaningful preoperative planning by gastric surgeons.

Keywords

Gastric carcinoma - lymph node metastasis - artificial neural network - prognostic factors - preoperative diagnosis - QUEEN

PDF (453 kb)

References

References
1 Siewert JR, Bottcher K, Roder JD. et al Prognostic relevance of systematic lymph node dissection in gastric carcinoma. German Gastric Carcinoma Study Group. Brit J Surg 1993; 80 (08) 1015-8.
2 Siewert JR, Bottcher K, Stein HJ, Roder JD. Relevant prognostic factors in gastric cancer: ten-year results of the German Gastric Cancer Study. Ann Surg 1998; 228 (04) 449-61.
3 de Manzoni G, Verlato G, Guglielmi A. et al Prognostic significance of lymph node dissection in gastric cancer. Brit J Surg 1996; 83 (11) 1604-7.
4 Ziegler K, Sanft C, Zimmer T. et al Comparison of computed tomography, endosonography, an intraoperative assessment in TN staging of gastric carcinoma. GUT 1993; 34: 604-10.
5 Akahoshi K, Chijiwa Y, Hamada S. et al Pretreatment staging of endoscopically early gastric cancer with a 15 MHz ultrasound catheder probe. Gastrointest Endosc 1998; 48: 470-6.
6 Davies J, Chalmers AG, Sue-Ling HM. et al Spiral computed tomography and operative staging of gastric carcinoma: a comparison with histopathological staging. GUT 1997; 41: 314-9.
7 Kim AY, Han JK, Seong CK, Kim TK, Choi BI. MRI in staging advanced gastric cancer: is it useful compared with spiral CT?. J Comput Assist Tomogr 2000; 24: 3389-94.
8 Mönig S, Zirbes TK, Schröder W. et al Staging of gastric cancer – correlation of lymph node size and metastatic infiltration. Am J Roentgenol 1999; 173: 365
9 Kampschöer GHM, Maruyama K, van de Velde CJH. et al Computer analysis in making preoperative decisions: a rational approach to lymph node dissection in gastric cancer patients. Br J Surg 1989; 76: 905-8.
10 Böhm BW. A spiral model of software development and enhancement. Computer 1988; 5: 604-9.
11 Waschulzik T. Anwendungsorientierte Entwicklung vorwärtsgerichteter künstlicher Neuronaler Netze mit überwachtem Lernen – Dissertation, TU München. 1999
12 Waschulzik T, Brauer W, Castedello T, Henery B. Quality Assured Efficient Engineering of Feedforward Neural Networks with Supervised Learning (QUEEN) Evaluated with the “Pima Indians Diabetes Database“. IEEE-INNS-ENNS International Joint Conference on Neural Networks (IJCNN’00)-Volume 4. 2000
13 Rosenblatt F. The perceptron: a probabilistic model for information storage and organization in the brain. Psychological Review 1958; 65: 386-408.
14 Frohn H, Geiger H, Singer W. A Self-Organizing Neural Network Sharing Features of the Mammalian Visual System. Biol Cybernetics 1987; 55: 333-43.
15 Hebb DO. The First Stage of Perception: Growth of the Assembly. In: The Organization of Behavior John Wiley & Sons; New York: 1949: 256-79.
16 Widrow B, Hoff ME. Adaptive switching circuits. IRE WESCON Convention Record. New York, IRE: 1960: 96-104.
17 Sasako M, Mc Culloch P, Kinoshita T, Maruyama K. New Method to evaluate the therapeutic value of lymph node dissection for gastric cancer. Br J Surg 1995; 82: 346-51.
18 Maruyama K. The Most Important Prognostic Factors for Gastric Cancer Patients – A Study Using Univariate and Multivariate Analyses. Scand Gastroenterol 1987; 22: 63-8.
19 Borrmann R. Geschwülste des Magens. In Henke FU, Lubarsch O. eds Handbuch der speziellen Pathologischen Anatomie und Histologie. Berlin: Springer Verlag; 1926
20 Bollschweiler E, Böttcher K, Hölscher AH. et al Preoperative assessment of lymph node metastasis in patients with gastric cancer: evaluation of the Maruyama computer program. Br JSurg 1992; 79: 156-60.
21 Guadagni S, de Manzoni G, Catarci M. et al Evaluation of the Maruyama computer program accuracy for preoperative estimation of lymph node metastases from gastric cancer. W J Surg 2000; 24 (12) 1550-8.
22 Japanese Gastric Cancer Association Japanese Classification of Gastric Carcinoma – 2nd English Edition. Gastric Cancer. 1998; 1: 10-24.
23 Droste K, Bollschweiler E, Waschulzik T. et al Prediction of lymph node metastasis in gastric cancer patients with neural networks. Cancer Letters 1996; 109: 141-8.
24 Baxt W. Application of artificial neural networks to clinical medicine. Lancet 1995; 346: 1135-8.
25 Schwarzer G, Vach W, Schumacher M. On the misuses of artificial neural networks for prognostic and diagnostic classification in oncology. Statist Med 2000; 19: 541-61.
26 Rodvold DM. Validation and regulation of Medical Neural Networks. Molecular Urology 2001; 4: 141-5.
27 Jordanov I, Brown R. Optimal design of connectivity in neural network training. Biomed Sci Instrum 2000; 36: 27-32.
28 Park J, Sandberg I W. Universal approximation using radial-basis-function networks. Neural Computation 1991; 3: 246-57.
29 Waschulzik T, Brauer W, Förster M. et al Quality Assurance and Increased Efficiency in Medical Projects with Neural Networks by using a Structured Development Method for Feedforward Neural Networks (SDNN). In Barahona P, Stefanelli M, Wyatt J. (eds.). Artificial Intelligence in Medicine, 5th Conference on Artificial Intelligence in Medicine Europe, AIME ’95 Pavia, Italy, Proceedings (1995)
30 Hölscher A, Bollschweiler E, Metzger R, Mönig S. D2 lymphadenectomy superior to d1 lymphadenectomy in gastric cancer surgery. Zentralbl Chir 2003; 128 (10) 786-7.
31 Bollschweiler E, Mönig S, Hölscher AH. Lymphknotenmetastasierung – Kann man sie vorhersagen?. Onkologe 2001; 7: 604-9.
32 Hunerbein M, Ulmer C, Handke T, Schlag PM. Endosonography of upper gastrointestinal tract cancer on demand using miniprobes or endoscopic ultrasound. Surgical Endoscopy (online) 2003; 17 (04) 615-9.